QuadSPI STM

AN4760
Application note
Quad-SPI interface on STM32 microcontrollers and
microprocessors
Introduction
In order to manage a wide range of multimedia, richer graphics and other data-intensive
content, embedded applications evolve to offer more sophisticated features. These
sophisticated features require extra demands on the often limited micocontroller (MCU) and
microprocessor (MPU) on-chip memory.
The STM32 MCUs and MPUs will be referred to as STM32 devices in this document. The
devices that are concerned are listed in Table 1: Applicable products
External parallel memories are used to extend the STM32 devices on-chip memory and
solve the memory size limitation. Usually this action compromises an increase in the pin
count and implies a more complex design.
To face these requirements, the STM32 devices embed an external memory interface
named Quad-SPI (see more details on Table 2 on page 9). This interface allows the
connection of external compact-footprint Quad-SPI high-speed memories.This Quad-SPI
interface is used for data storage such as images, icons, or for code execution.
This application note describes the Quad-SPI interface on the STM32 devices and explains
how to use the module to configure, program, and read external Quad-SPI memory. It
describes some typical use cases to use the Quad-SPI interface based on some software
examples from the STM32Cube firmware package and from the STM32F7 Series
application notes.
For additional more detailed information about the products listed in the table below, refer to
the corresponding datasheets and reference manuals available from the STMicroelectronics
web site www.st.com.
Table 1. Applicable products

Type Products, lines and series
STM32F7 Series, STM32L4 Series

STM32F412, STM32F413/423, STM32F446, STM32F469/479,
STM32H743/753, STM32H750 Value line
Microcontrollers STM32L4R5/S5, STM32L4R7/S7, STM32L4R9/S9
STM32WB55CC, STM32WB55CE, STM32WB55CG, STM32WB55RC,
STM32WB55RE, STM32WB55RG, STM32WB55VC, STM32WB55VE,
STM32WB55VG, STM32WB35CC, STM32WB35CE, STM32WB35CZ
Microprocessors STM32MP151x, STM32MP153x, STM32MP157x devices
April 2020 AN4760 Rev 3 1/95

www.st.com 1
Contents AN4760
Contents
1 General information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1 QUADSPI availability and features across STM32 families . . . . . . . . . . . . 8
2.2 Quad-SPI benefits against classic SPI and parallel interfaces . . . . . . . . . 10
2.2.1 Main benefits of STM32 embedded Quad-SPI interface . . . . . . . . . . . . 10
2.3 QUADSPI in a smart architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11
2.3.1 System architecture: STM32L4 Series . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.2 System architecture: STM32F4 Series . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.3 System architecture: STM32F7 Series . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.4 System architecture: STM32H7 Series . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.5 System architecture: STM32WB35xx and STM32WB55xx devices . . . 16
3 Quad-SPI interface description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3.1 Flexible frame format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.1 Instruction phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.2 Address phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.3 Alternate-byte phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.4 Dummy-cycle phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.5 Data phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Multiple hardware-configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.1 Single-SPI mode (classic SPI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.2 Dual-SPI mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2.3 Quad-SPI mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.4 Dual-Flash memory mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.5 DDR and SDR mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 Three operating modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3.1 Indirect mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3.2 Status-flag polling mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.3 Memory-mapped mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Special features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4.1 Send instruction only-once (SIOO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4.2 Delayed data sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4.3 Timeout counter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2/95 AN4760 Rev 3

AN4760 Contents
3.4.4 Additional status bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.4.5 Busy bit and abort functionality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.4.6 4-byte address mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.7 QUADSPI and delay block in STM32H7 Series . . . . . . . . . . . . . . . . . . 35
3.5 Interrupts and DMA usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5.1 Interrupts usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5.2 DMA usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.6 Low-power modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4 QUADSPI configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1 GPIOs configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1.1 GPIOs configuration using STM32CubeMX tool . . . . . . . . . . . . . . . . . . 40
4.2 QUADSPI peripheral configuration and clock . . . . . . . . . . . . . . . . . . . . . 43
4.2.1 QUADSPI peripheral configuration (QUADSPI_CR register) . . . . . . . . 43
4.2.2 Quad-SPI Flash memory parameters configuration
(QUADSPI_DCR register) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.2.3 QUADSPI and MPU configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2.4 Quad-SPI memory device configuration . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2.5 Starting a communication (QUADSPI_CCR register) . . . . . . . . . . . . . . 47
4.3 Hardware considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3.1 Pull-up resistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3.2 Good PCB design allows maximum QUADSPI speed . . . . . . . . . . . . . 48
4.3.3 Chip-select high time (CSHT) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3.4 CKMODE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.5 Some considerations when using QUADSPI in classical SPI mode . . . 49
5 Programming Quad-SPI Flash memory . . . . . . . . . . . . . . . . . . . . . . . . 50

5.1 Programming code or data for an end application . . . . . . . . . . . . . . . . . . 50
5.1.1 Programming Quad-SPI Flash memory using
the STM32 ST-LINK utility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.1.2 Programming Quad-SPI Flash memory using IDE . . . . . . . . . . . . . . . . 55
5.2 Storing and erasing data on the fly during running application . . . . . . . . 59
5.2.1 Storing data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2.2 Erasing data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6 QUADSPI application examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

6.1 Reading data from Quad-SPI memory: graphical application . . . . . . . . . 62
AN4760 Rev 3 3/95

4
Contents AN4760
6.1.1 Frame buffer content generation from Quad-SPI memory . . . . . . . . . . . 62

6.1.2 Displaying images directly from the Quad-SPI memory . . . . . . . . . . . . 65
6.2 Executing from external Quad-SPI memory: extend internal memory size 67
6.2.1 Configuring Quad-SPI in Memory-mapped mode during system
initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.2.2 Placing application code in external Quad-SPI memory . . . . . . . . . . . . 73
6.3 Storing (programming) data on the fly during a running application . . . . . 79
6.3.1 QUADSPI indirect write: programming Quad-SPI memory using DMA . 79
6.3.2 QUADSPI indirect write: programming Quad-SPI memory using
interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.4 Erasing-data example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.5 Hardware implementation example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7 Performance and power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

7.1 How to get the best performances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
7.1.1 Write performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
7.1.2 Read performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
7.2 Decreasing power consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.2.1 Use timeout counter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.2.2 Put the Quad-SPI memory in Deep power-down mode . . . . . . . . . . . . . 90
7.2.3 Quad-SPI Flash memories supporting DPD mode . . . . . . . . . . . . . . . . 91
8 Supported devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
10 Revision history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4/95 AN4760 Rev 3

AN4760 List of tables
List of tables
Table 1. Applicable products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Table 2. QUADSPI availability and features across STM32 families . . . . . . . . . . . . . . . . . . . . . . . . . 9
Table 3. Benefits of using STM32 Quad-SPI interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Table 4. Instruction phase configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Table 5. Address-phase configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Table 6. Alternate-byte phase configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Table 7. Data phase configuration versus Quad-SPI functional modes . . . . . . . . . . . . . . . . . . . . . . 22
Table 8. Hardware configurations versus used GPIO number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Table 9. Dual-Flash memory hardware configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Table 10. Additional status bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Table 11. BUSY bit reset in different Quad-SPI modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Table 12. Address mode versus maximum addressable memory space . . . . . . . . . . . . . . . . . . . . . . 35
Table 13. QUADSPI interrupts summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Table 14. DMA requests mapping and transfer directions versus STM32 series . . . . . . . . . . . . . . . . 37
Table 15. Execution performances versus configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Table 16. Different STM32 boards embedding Quad-SPI Flash memory . . . . . . . . . . . . . . . . . . . . . 85
Table 17. Document revision history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
AN4760 Rev 3 5/95

5
List of figures AN4760
List of figures
Figure 1. System architecture: STM32L4 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Figure 2. System architecture: STM32F4 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Figure 3. System architecture: STM32F7 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Figure 4. System architecture: STM32H7 Series. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Figure 5. System architecture:STM32WB35xx and STM32WB55xx . . . . . . . . . . . . . . . . . . . . . . . . . 16
Figure 6. Reading sequence in quad I/O SDR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Figure 7. Alternate-byte phase: sending a nibble in dual-SPI mode . . . . . . . . . . . . . . . . . . . . . . . . . 20
Figure 8. Dummy-cycle: IO2 maintained low and IO3 maintained high by hardware . . . . . . . . . . . . 21
Figure 9. Hardware configuration: Single-SPI mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Figure 10. Hardware configuration: dual-SPI mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Figure 11. Hardware configuration: Quad-SPI mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Figure 12. Read sequence in dual-Flash memory Quad I/O SDR mode . . . . . . . . . . . . . . . . . . . . . . . 26
Figure 13. Executing non-sequential code from Quad-SPI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Figure 14. Executing non-sequential code from QUADSPI with SIOO enabled . . . . . . . . . . . . . . . . . 32
Figure 15. QUADSPI and delay block . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Figure 16. QUADSPI and master DMA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Figure 17. STM32CubeMX: QUADSPI GPIOs configuration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Figure 18. STM32CubeMX: PF8 pin configuration to QUADSPI_BK1_IO0 alternate function . . . . . . 41
Figure 19. STM32CubeMX: Dual-Flash memory QUADSPI with chip-select 1 configuration . . . . . . . 42
Figure 20. STM32CubeMX: enabling QUADSPI global interrupt. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Figure 21. QUADSPI clock configuration on QUADSPI_CR register. . . . . . . . . . . . . . . . . . . . . . . . . . 43
Figure 22. STM32CubeMX: quadspi_ker_ck source clock configuration in
STM32H7 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Figure 23. STM32CubeMX: quadspi_ker_ck source clock selection in
STM32H7 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Figure 24. STM32CubeMX: QUADSPI peripheral configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Figure 25. Write enable sequence (command 0x06) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Figure 26. Connecting chip-select to a pull-up resistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Figure 27. Chip select high time: CSHT = two clock cycles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Figure 28. QUADSPI in classical SPI mode frame example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Figure 29. Programming Quad-SPI memory through debug interface. . . . . . . . . . . . . . . . . . . . . . . . . 51
Figure 30. STM32 ST-LINK utility: adding Quad-SPI Flash memory loader . . . . . . . . . . . . . . . . . . . . 52
Figure 31. STM32 ST-LINK utility: selecting Quad-SPI Flash memory loader. . . . . . . . . . . . . . . . . . . 52
Figure 32. STM32 ST-LINK utility: error message . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Figure 33. STM32 ST-LINK utility: programming Quad-SPI Flash memory. . . . . . . . . . . . . . . . . . . . . 53
Figure 34. STM32 ST-LINK utility: selecting HEX file for programming. . . . . . . . . . . . . . . . . . . . . . . . 53
Figure 35. STM32 ST-LINK utility: erasing sectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Figure 36. Adding Quad-SPI Flash memory loader to Keil MDK-ARM project . . . . . . . . . . . . . . . . . . 56
Figure 37. Adding Quad-SPI Flash memory loader to Keil MDK-ARM project . . . . . . . . . . . . . . . . . . 57
Figure 38. Selecting Quad-SPI Flash memory programming algorithm. . . . . . . . . . . . . . . . . . . . . . . . 57
Figure 39. Quad-SPI Flash memory loader programming algorithm configuration . . . . . . . . . . . . . . . 58
Figure 40. Quad I/O page program sequence (command 0x38) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Figure 41. Read status register sequence (command 0x05) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Figure 42. Sector erase sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Figure 43. Example: full chip-erase sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Figure 44. QUADSPI usage in a graphical application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Figure 45. DMA2D reading images from Quad-SPI to build frame buffer content . . . . . . . . . . . . . . . . 65
Figure 46. LTDC reading an image directly from Quad-SPI memory . . . . . . . . . . . . . . . . . . . . . . . . . 66
6/95 AN4760 Rev 3

AN4760 List of figures
Figure 47. Project configurations: executing code from Quad-SPI Flash memory . . . . . . . . . . . . . . . 68
Figure 48. Changing QUADSPI configuration in the project settings. . . . . . . . . . . . . . . . . . . . . . . . . . 69
Figure 49. Quad-SPI Flash memory connection in STM32756-EVAL board . . . . . . . . . . . . . . . . . . . . 70
Figure 50. 6_1-Quad-SPI_rwRAM-DTCM project configuration: code and data in
Quad-SPI memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Figure 51. 6_2-Quad-SPI_rwRAM-DTCM project configuration: only code in Quad-SPI memory . . . 76
Figure 52. Indirect write mode: programming Quad-SPI memory using DMA . . . . . . . . . . . . . . . . . . . 80
Figure 53. Indirect write mode: programming Quad-SPI memory using interrupt . . . . . . . . . . . . . . . . 82
Figure 54. Quad-SPI memory connection on the STM32F746G-DISCO discovery board . . . . . . . . . 86
Figure 55. Quad-SPI memory connection on the STM32L476G-EVAL board. . . . . . . . . . . . . . . . . . . 86
Figure 56. Deep power-down (DPD) sequence (command B9). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Figure 57. Release from deep power-down (RDP) sequence (command AB) . . . . . . . . . . . . . . . . . . 91
AN4760 Rev 3 7/95

7
General information AN4760
1 General information
This document applies to STM32 Arm®-based(a) microcontrollers and microprocessors.
2 Overview
The Quad-SPI is a serial interface that allows the communication on four data lines between
a host (STM32) and an external Quad-SPI memory. The QUADSPI supports the traditional
SPI (serial peripheral interface) as well as the dual-SPI mode which allows to communicate
on two lines. QUADSPI uses up to six lines in quad mode: one line for chip select, one line
for clock and four lines for data in and data out.
This interface is integrated on the STM32 devices to fit memory-hungry applications, to
simplify PCB (printed circuit board) designs and to reduce costs.
2.1 QUADSPI availability and features across STM32 families

All STM32 devices shown in the table below have mainly the same QUADSPI features.
a. Arm is a registered trademark of Arm Limited (or its subsidiaries) in the US and/or elsewhere.
8/95 AN4760 Rev 3

AN4760 Overview
Table 2. QUADSPI availability and features across STM32 families

Maximum speed (MHz)(1) Max addressable space (2)
Products Dual- FIFO size
Flash
(byte) Memory
SDR DDR memory Indirect mode
mapped
STM32F412 line
100 80
STM32F413/423 line(3)
STM32F446 line(4) 60
90
STM32F469/479 line
STM32F730xx devices
STM32F7x2 line(4)
STM32F750xx Yes 32
STM32F7x3 80
STM32F7x5 108
STM32F7x6
STM32F7x7
STM32F7x8
STM32F7x9
STM32H743/753
133 100
STM32H750 Value line
STM32L471xx
STM32L412xx
STM32L422xx 256 Mbytes 4 Gbytes
STM32L432xx
No
STM32L442xx
STM32L475xx
STM32L476xx
STM32L486xx
48
STM32L431xx 60 16
STM32L451xx
STM32L452xx
STM32L462xx Yes
STM32L4x3(5)
STM32L496xx
STM32L4A6xx
STM32WB35xx
50 No
STM32WB55xx
STM32L4R5/S5
STM32L4R7/S7 86 60 Yes
STM32L4R9/S9(6) 32
STM32MP1 166 90 Yes

1. Maximum QUADSPI speed from datasheet. For more details on the QUADSPI maximum speed refer to the relevant device
datasheet.
2. 32-bits address mode should be used to reach 256 Mbytes in Memory-mapped mode and 4 Gbytes in Indirect mode.
3. UFQFPN48 does not support Quad-SPI.
4. LQFP64 supports only Bank1 and Single-SPI/Dual-SPI only.
5. For this set of products, Dual-Flash mode is supported only with LQFP100 and UFBGA100 packages.
AN4760 Rev 3 9/95

94
Overview AN4760
6. This set of products contains two Octo-SPI interfaces, each one of them can connect one or two Quad-SPI memories with
Single-Flash or Dual-Flash modes.
2.2 Quad-SPI benefits against classic SPI and parallel interfaces

The Quad-SPI brings more performance in terms of throughput compared to classical SPI.
The classical SPI uses only one data line while the Quad-SPI uses four data lines which
multiplies the data throughput by almost four times.
Compared to FMC (flexible memory interface) and other parallel interfaces, Quad-SPI
permits the connection of a lower cost external Flash memory to small packages, reducing
the PCB area, simplifying the PCB design and reducing the GPIOs (general-purpose
input/output) usage. In Quad-SPI mode, only six GPIOs are used: four lines for data plus
one line for clock and another for chip select. In Dual-Flash Quad-SPI mode only 10 GPIOs
are used, amongst which eight lines are for data.
2.2.1 Main benefits of STM32 embedded Quad-SPI interface

The table below summarizes the major advantages of using STM32 embedded Quad-SPI
interface:
Table 3. Benefits of using STM32 Quad-SPI interface

Benefits Comments
Supports single, dual and Quad-SPI memories.

Low pin-count Uses six pins in Quad-SPI mode and four pins for single or dual-SPI.
Saves GPIOs to be used for other purposes.
Easier PCB design Allows easier and faster PCB design thanks to a reduced pin count.
Save space for smaller Can be used in small size applications due to small footprint Quad-SPI
size applications memories.
Easier and faster design permits a lower development cost.
Lower PCB cost, as it is possible to reduce PCB layers due to low pin-
Save cost
count.
Low cost memory solution.
Extends limited on-chip Flash memory allowing Quad-SPI memory to be
seen as an internal memory.
Allows code execution (XIP mode) from Quad-SPI Flash memory.
Executable
Supports SIOO mode also named Continuous read mode by some
memory manufacturers (see Section 3.4.1: Send instruction only-once
(SIOO) on page 32) for higher execution performance.
Memory-mapped mode allows Quad-SPI memory to be accessed
autonomously by any AHB (advanced high-performance) or AXI
(advanced extensible Interface protocol) master.
Extended size for data
32-bits address mode enables the possibility to address up to four
storage
Gbytes Quad-SPI memory size.
Dual-Flash memory mode enables the use of two Quad-SPI Flash
memories to double storage size(1).
10/95 AN4760 Rev 3

AN4760 Overview
Table 3. Benefits of using STM32 Quad-SPI interface (continued)

Benefits Comments
Throughput is multiplied by four versus traditional SPI.

The DDR mode doubles throughput.
High performances
The Dual-Flash memory mode doubles throughput.
Perfect for graphical applications.
There are volatile Quad-SPI SRAM (static random-access memory)
available from Microchip, ON Semiconductor and others.
Multiple memory solutions
Available non-volatile Quad-SPI Flash memories.
NOR, NAND.
Supports any Quad-SPI
Its fully configurable and flexible frame format permits to support almost
memories available in the
all Quad-SPI devices available on the market.
market
Spansion, Windbond, Micron, Macronix, ONSemiconductors, Cypress,
Growing amount of APmemory and ISSI among others.
manufacturers Huge investment on higher densities Quad-SPI Flash memories such as
NAND.
1. 4 Gbytes maximal size can be reached with the 32-bits address mode.
2.3 QUADSPI in a smart architecture

The Quad-SPI interface is mapped on a dedicated layer on AHB allowing it to be accessible
as an internal memory thanks to the Memory-mapped mode. In addition, the QUADSPI is
integrated in a smart architecture which allows the following features:
• Masters to access the external Quad-SPI memory without any CPU intervention.
• Masters to read data from Quad-SPI memory even in Sleep mode when the CPU is
stopped thanks to the STM32 smart architecture.
• CPU as a master can access QUADSPI and execute code from the memory.
• GP DMA to do transfer from Quad-SPI to other internal or external memories.
• Graphical DMA2D to directly build RAM video frames using Quad-SPI Flash.
AN4760 Rev 3 11/95

94
Overview AN4760
2.3.1 System architecture: STM32L4 Series

The STM32L4 Series system architecture consists mainly of a 32-bit multilayer AHB bus
matrix that interconnects multiple masters to multiple slaves.
The QUADSPI can be accessed by relevant masters like the Arm® Cortex®-M4 either
through
S-Bus or through I-bus and D-bus when remap is enabled. QUADSPI is also accessible by
DMA1 and DMA2.
Enabling physical remap over I-bus and D-bus boosts execution performances for the
Cortex®-M4.
The access to the QUADSPI can be either a registers access or a memory-mapped region
access:
• The registers access can be done by the Cortex®-M4 for registers configuration or data
transfer. The register access can be done also by the DMA1 and DMA2 for data
transfer.
• The memory mapped region access can be done by the Cortex®-M4 for code and data
fetch. The memory-mapped region can also be accessed by the DMA1, DMA2 and
DMA2D for data transfer.
The figure below shows a QUADSPI interconnection in the STM32L4 Series system.
Note: DMA2D is available only in STM32L496xx and STM32L4A6xx devices.
Figure 1. System architecture: STM32L4 Series

2
Cortex-M4 DMA1 DMA2 DMA2D

D-Bus
S-Bus
I-Bus
ICode
ACCEL
Flash
Masters accessing
DCode memory QUADSPI
Quad-SPI interface
SRAM1
32-bit AHB bus
SRAM2 QUADSPI registers

access
AHB1 peripheral QUADSPI memory-

mapped region access
AHB2 peripheral Bus mutliplexer
4 4 3
FMC
4 4 Registers
1
Memory-mapped region
1 For STM32L471xx, STM32L475xx, STM32L476xx and STM32L486xx devices, QUADSPI and FMC share the same AHB bus on the bus matrix
2 DMA2D is only available on STM32L496xx and STM32L4A6xx devices
3 FMC is available only on STM32L47xxx and STM32L4x6xx devices
4 When remapped
12/95 AN4760 Rev 3

AN4760 Overview
2.3.2 System architecture: STM32F4 Series

The STM32F4 Series system architecture consists mainly of a 32-bit multilayer AHB bus
matrix that interconnects multiple masters to multiple slaves (refer to cover page for detail
on applicable products).
The external Quad-SPI memory can be accessed by the Cortex®-M4 through the S-bus.
The QUADSPI is also accessible by all the masters on the AHB bus matrix such as DMA1,
DMA2, USB OTG HS, MAC Ethernet, LTDC and DMA2D. This accessibility enables an
efficient data transfer (like images for graphical applications).
access:
• The registers access can be done by the Cortex®-M4 through S-Bus for registers
configuration and data transfer. The register access can be done also GP DMA2 for
data transfer.
• The memory-mapped region access can be done by the Cortex®-M4 through S-Bus for
code and data fetch.The memory-mapped region access can be done also by the
GP DMA1, GP DMA2, MAC Ethernet, USB OTG HS, LTDC and DMA2D for data
transfer.
The figure below shows a QUADSPI interconnection in the STM32F4 Series system.
Note: For MAC Ethernet, USB OTG HS, LTDC and DMA2D refer to the applicable product.
Figure 2. System architecture: STM32F4 Series

2 3 2 2
Ethernet
OTG HS
64-Kbyte 2
Chrom-
DMA1
DMA2
LDTC
MAC
USB
GP
GP
CCM data Cortex-M4 ART

RAM (DMA2D)
D-bus
S-bus
DMA_P2
DMA_PI
I-bus
DMA_MEM1
DMA_MEM2
Flash
ART
Masters accessing memory

QUADSPI
Quad-SPI interface SRAM1

SRAM2
32-bit AHB bus
SRAM3 APB1
QUADSPI registers
access AHB1 peripheral APB2
QUADSPI AHB2 peripheral
memory-mapped
region access FMC
Bus mutliplexer Registers
1
Bus matrix-S Memory-mapped region
1 For STM32F412, STM32F413/423 and STM32F446 lines, QUADSPI and FMC share the same AHB bus on the bus matrix
2 Available only on STM32F469/479 line devices
3 USB OTG HS is available only in the STM32F446 and STM32F469/479 lines
AN4760 Rev 3 13/95

94
Overview AN4760
2.3.3 System architecture: STM32F7 Series

The main system architecture is based on two subsystems, an AXI (advanced extensible
interface) to multi AHB bridge converting AXI4 protocol to AHB-Lite protocol and a multi-
AHB bus matrix.
The multi AHB bus matrix interconnects multiple masters and multiple slaves. There are four
AXI bus accesses; the QUADSPI is accessible through the second access. This access
allows the Cortex®-M7 to perform a memory-mapped region access in order to fetch code or
data. This access also allows the Cortex®-M7 to perform a register access for QUADSPI
registers configuration or for data transfer.
The QUADSPI is mapped on a dedicated layer on the AHB Bus matrix allowing the
Cortex®-M7 to benefit from L1-Cache when accessing the cached data with 0-wait states.
QUADSPI is also accessible by all masters on AHB bus matrix. Registers accesses can be
performed by GP DMA2 for data transfer. Memory-mapped region access can be performed
by GP DMA1, MAC Ethernet, USB OTG HS, LTDC and DMA2D. This accessibility enables
an efficient data transfer (like images for graphical applications).
The following figure shows the QUADSPI interconnection in the STM32F7 Series system.
Note: For MAC Ethernet, USB OTG HS, LTDC and DMA2D refer to the applicable product.
Figure 3. System architecture: STM32F7 Series

DTCM
ITCM
AHBS
1 1 1
OTG HS
Ethernet
DMA2D
Arm Cortex -M7

DMA1
DMA2
LDTC
USB
MAC
GP
GP
L1-cache DTCM RAM

ITCM RAM
DMA_MEM1
DMA_MEM2
DMA_P2
AXIM AHBP
DMA_P1
AXI to muti AHB

APB1
APB2
ART
Flash
memory
64-bit AHB
64-bit bus matrix Masters accessing
QUADSPI
SRAM1
Quad-SPI interface
SRAM2
32-bit AHB bus
AHB1 peripheral
64-bit AHB bus
AHB2 peripheral
FMC QUADSPI registers
access
Registers
Memory-mapped region QUADSPI
memory-mapped
region access
32-bit bus matrix-S
Bus mutliplexer
1 Mac Ethernet , LCD-TFT and DMA2D are not available on STM32F72xxx and STM32F73xxx devices.
14/95 AN4760 Rev 3

AN4760 Overview
2.3.4 System architecture: STM32H7 Series

The main system architecture is based on three domains: D1,D2 and D3. Each domain
contains a bus matrix that allows a connection between multiple masters and multiple
slaves. A 64-bit AXI bus matrix for domain D1 and a 32-bit AHB bus matrix for each of the
domains D2 and D3.
The three domains are connected to each other with the interdomains AHB buses which
allow masters from a certain domain to access slaves from an other domain.
The QUADSPI is connected to the D1 domain and can be accessed through:
• A 64-bit AXI bus connected directly to the AXI bus matrix. It allows multiple masters to
perform a memory-mapped region access for code and data fetch from D1 domain like
the Cortex®-M7. It allows also data transfer from D1 domain (SDMMC1, MDMA,
DMA2D and LTDC) and from the D2 domain (DMA1 and DMA2).
• A 32-bit AHB bus accessible through AHB3. It allows the Cortex®-M7 and the MDMA to
perform a register access for data transfer or registers configuration.
The following figure shows the QUADSPI interconnection in the STM32H7 Series system.
Figure 4. System architecture: STM32H7 Series

AHBS
SDMMC2
USB HS1
USB HS2
Ethernet
MAC
Cortex-M7 ITCM
D$ DMA1 DMA2
I$
DTCM
DMA1_PERIPH
DMA2_PERIPH
AHBP
DMA1_MEM
DMA2_MEM
SDMMC1 MDMA DMA2D LTDC
D1-to-D2 AHB
APB3 SRAM1
Registers
AHB3 SRAM2
FLASH A SRAM3
1
FLASH B AHB2
AXI SRAM AHB1
Memory-mapped region APB1
FMC APB2
64-bit AXI bus matrix

D1-to-D AHB
D1 domain
32-bit AHB bus matrix

D2-to-D1 AHB D2 domain
D2-to-D3 AHB
BDMA
Quad-SPI interface MDMA
Masters not having Masters accessing

access to QUADSPI QUADSPI
AHB4 APB4
AXI QUADSPI registers access
AHB
SRAM4
QUADSPI memory-mapped region access
APB
TCM
32-bit AHB bus matrix
64-bit bus width 32-bit bus width D3 domain Backup SRAM
Bus mutliplexer
1
Mac Ethernet , LCD-TFT and DMA2D are not available on STM32F72x and STM32F73x devices
AN4760 Rev 3 15/95

94
Overview AN4760
2.3.5 System architecture: STM32WB35xx and STM32WB55xx devices

The STM32WB35xx, STM32WB55xx devices system architecture consists mainly of a 32-
bit multilayer AHB bus matrix that interconnects multiple masters and slaves. The QUADSPI
is mapped on a dedicated layer on the AHB bus matrix.
The QUADSPI is accessed by relevant masters like the Cortex®-M4 either through S-Bus or
through I-bus and D-bus when remap is enabled. The QUADSPI is also accessible by
DMA1 and DMA2.
Enabling physical remap over I-bus and D-bus boosts execution performances for the
Cortex®-M4.
access:
• The registers access can be done by the Cortex®-M4 for registers configuration or data
transfer. The register access can also by done by DMA1and DMA2 for data transfer.
• The memory-mapped region access can be done by the Cortex®-M4 for code and data
fetch and also by the DMA1 and DMA2 for data transfer.
The following figure shows the QUADSPI interconnection in the STM32WB35xx and
STM32WB55xx devices system.
Figure 5. System architecture:STM32WB35xx and STM32WB55xx system
CPU1 CPU2
DMA1
DMA2
Radio
Arm Cortex-M4 Arm Cortex-M0+

D-bus
I-bus
S-bus
ICode
DCode CFI Flash
SCode arbiter memory
SRAM1
SRAM2
AHB1 peripheral
AHB2 peripheral
Registers
1 1
Memory-mapped region
AHB4
AHB5
Bus matrix
32-bit AHB bus

Quad-SPI interface
QUADSPI registers access
Masters not having access to
QUADSPI memory-mapped region access QUADSPI
Bus mutliplexer Masters accessing QUADSPI
1 When remapped
16/95 AN4760 Rev 3

AN4760 Quad-SPI interface description
3 Quad-SPI interface description
3.1 Flexible frame format

The Quad-SPI interface provides a fully programmable frame composed of five phases
where each phase is fully configurable, allowing it to be configured separately in terms of
length and number of lines.
The frame format can be configured only in Indirect mode or Memory-mapped mode but not
in Status-flag polling mode. The figure below shows a reading of the sequence in Quad I/O
SDR mode.
Figure 6. Reading sequence in quad I/O SDR
Instruction Address Alt Dummy Data
nCS
SCLK
IO 0 4 0 4 0 4 0 4 0 4 0 4 0 4 0
IO 1 5 1 5 1 5 1 5 1 5 1 5 1 5 1
IO 2 6 2 6 2 6 2 6 2 6 2 6 2 6 2
IO 3 7 3 7 3 7 3 7 3 7 3 7 3 7 3
A23-16 A15-8 A7-0 M7-0 Byte 1 Byte 2
Only 2 cycles
command IO switch from
output to input
MSv41107V1
3.1.1 Instruction phase

In this phase a command (8-bits instruction) is sent to the Flash memory, specifying the type
of operation to be performed.This command is fully configurable allowing to send any value.
The user can simply write the desired command to be sent in the INSTRUCTION field of the
QUADSPI_CCR[7:0] register.
Depending on the software and the hardware configurations, the instruction can be sent
over one, two or four lines. In some use cases where only the address is sent, the
instruction phase can be skipped. The following table summarizes the different
configurations for instruction phase.
Note: The DDR mode is not supported in this phase, so even if the DDR mode is enabled the
command is always sent in SDR mode.
AN4760 Rev 3 17/95

94
Quad-SPI interface description AN4760
Table 4. Instruction phase configurations

Indirect mode
Automatic-
Register configurations polling mode Command formats
Memory-
mapped mode
Instruction to be sent in INSTRUCTION

NA
QUADSPI_CCR[7:0] [7: 0]
No Instruction:
IMODE[1:0] = 00 The instruction phase is skipped
skipped
nCS
CLK
Instruction on 1 BK1_IO0 7 6 5 4 3 2 1 0 output
line: Single SPI IMODE[1:0] = 01
mode BK1_IO1 High-Z
BK1_IO2 High-Z
BK1_IO3 High-Z
MSv41108V1
nCS
Instruction phase CLK

Instruction on 2
QUADSPI_CCR[9:8]
lines: Dual SPI IMODE[1:0] = 10
BK1_IO0 6 4 2 0 Output
mode BK1_IO1 7 5 3 1 Output
BK1_IO2 High-Z
BK1_IO3 High-Z
MSv41109V1
nCS
CLK
Instruction on 4
BK1_IO0 4 0 Output
lines: Quad-SPI IMODE[1:0] = 11
mode BK1_IO1 5 1 Output
BK1_IO2 6 2 Output
BK1_IO3 7 3 Output MSv41110V1
18/95 AN4760 Rev 3

3.1.2 Address phase

In this phase an address is sent to the Flash memory, specifying the address of the data to
be read or written. The address phase is fully configurable allowing to send one, two, three
or four bytes address. In Indirect mode and Automatic-polling mode, the user can simply
write the desired address in the QUADSPI_AR register.
Depending on the software and the hardware configurations, the address can be sent over
one, two or four lines. In some use cases where the address is not needed such as in mass-
erase operation, the address phase can be skipped
The table below summarizes different address phase configurations.
Table 5. Address-phase configurations

Automatic-
Indirect Memory-mapped
Register configurations polling
mode mode
mode
Address is given
directly via the AHB
Address to be sent QUADSPI_AR[31:0] ADDRESS[31:0] from any master on
the bus matrix like
Cortex® or DMA
1-byte ADSIZE[1:0] =00
Address size 2-byte ADSIZE[1:0] =01

QUADSPI_CCR[13:12] 3-byte ADSIZE[1:0] =10
4-byte ADSIZE[1:0] =11
No address: skipped ADMODE[1:0] =00
Address on 1 line:
ADMODE[1:0] =01
Single SPI mode
Address phase
QUADSPI_CCR[11:10] Address on 2 lines: ADMODE[1:0] =10
Dual SPI mode
Address on 4 lines:
ADMODE[1:0] =11
Quad SPI mode
Note: In Dual-Flash memory mode when DFM = 1 the address to be sent to Flash1 is exactly the
same address to be sent to Flash2.
3.1.3 Alternate-byte phase

This is an extra phase supported by the Quad-SPI interface offering more flexibility. It is
generally used for controlling the mode of operation; for instance 1-byte can be sent
continuously to keep the Quad-SPI device in an operating mode. This is supported for some
memory manufacturers such as Spansion, Micron and Macronix where an alternate byte is
sent continuously to keep the memory in execute-in-place mode.
The alternate-byte phase is fully configurable and permits to send one, two, three or four
bytes depending on the ABRSIZE [1:0] file configuration. The user can simply write the
desired alternate bytes in the QUADSPI_ABR register.
AN4760 Rev 3 19/95

94
Depending on the software and the hardware configurations, the alternate byte can be sent
over one, two or four lines. If not needed, the alternate-byte phase can be skipped.
The table below summarizes different alternate-byte phase configurations.
Table 6. Alternate-byte phase configurations

Automatic Memory-
Indirect
Register configuration -polling mapped
mode
mode mode
Alternate-byte to be sent QUADSPI_ABR QUADSPI_ABR

1-byte ABSIZE [1:0] =00
Alternate-byte size 2-byte ABSIZE [1:0] =01

QUADSPI_CCR[17:16] 3-byte ABSIZE [1:0] =10
4-byte ABSIZE [1:0] =11
No alternate-byte: skipped ABMODE [1:0] = 00
Alternate-byte on 1 line:
ABMODE [1:0] = 01
single SPI mode
Alternate-byte phase
QUADSPI_CCR[15:14] Alternate-byte on 2 lines:
ABMODE [1:0] = 10
dual SPI mode
Alternate-byte on 4 lines:
ABMODE [1:0] = 11
Quad SPI mode
Note: In Dual-Flash memory mode when DFM = 1, the alternate-byte to be sent to Flash1 is
exactly the same as the ones to be sent to Flash2.
Alternate-byte phase: sending a nibble in Dual-SPI mode

In some cases only one nibble needs to be sent at alternate-byte phase during two clock
cycles rather than a full byte during four clock cycles. For instance, when the Dual-SPI
mode is used and only two cycles are used for the alternate-byte phase.
The Quad I/O mode can be activated only for alternate-byte phase in order to send an
alternate byte where the nibble is sent over IO0 and IO1 while the other nibble have to be
sent only to keep IO2 low and IO3 high during alternate-byte phase.
Figure 7. Alternate-byte phase: sending a nibble in dual-SPI mode
Alternate-byte
nCS
Nibble to be sent 0010
CLK
BK1_IO0 0 0 Output
BK1_IO1 0 1 Output
BK1_IO2 0 0 nWP
BK1_IO3 1 1 nHOLD
MSv41111V1
20/95 AN4760 Rev 3

3.1.4 Dummy-cycle phase

The dummy-cycle phase is needed in some cases when operating at high clock
frequencies. This phase allows to ensure enough “turnaround” time for changing the data
signals from output mode to input mode.
The dummy phase is enabled by setting the number of dummy cycles in the DCYC[4:0] field
QUADSPI_CCR register. The number defined in DCYC[4:0] filed can reach 31 cycles and it
does not depend on the used hardware configuration.
Note: Either in SDR or DDR mode, one dummy represents always one QUADSPI clock cycle.
During this phase, if the QUADSPI hardware configuration is used and if communication
phases are either in Quad-SPI or Dual-SPI modes, IO2 is forced to ‘0’ to disable the “write
protect” function while IO3 is forced to ‘1’ to disable the “hold” function of the Quad-SPI
memory. This is fully managed by hardware (QUADSPI peripheral) so nothing needs to be
configured by the user.
Figure 8. Dummy-cycle: IO2 maintained low and IO3 maintained high by hardware
2 Dummy cycles
nCS
CLK
BK1_IO0 High-Z
BK1_IO1 High-Z
BK1_IO2 0 nWP
BK1_IO3 1 nHOLD
MSv41160V1
3.1.5 Data phase

In this phase; the data is sent or received from or to the Quad-SPI memory. The data phase
is fully configurable and permits to send, receive or both any number of bytes to or from the
Quad-SPI memory device.
In Indirect mode and in Automatic-polling mode, the number of bytes to be sent, received or
both is specified in the QUADSPI_DLR register.
In Indirect-write mode the data to be sent to the Flash memory must be written to the
QUADSPI_DR register, while in Indirect-read mode the data received from the Flash
memory is obtained by reading from the QUADSPI_DR register.
In Memory-mapped mode, the data can only be read from the memory device but not
written, then the data is accessed directly from the QUADSPI FIFO. All masters on the bus
matrix can read data from the Quad-SPI memory device as if it was an internal memory.
AN4760 Rev 3 21/95

94
Depending on the software and the hardware configurations, the data transfer can be done
in one, two or four lines. In some use cases where data is not needed such as erasing
operation, the data phase can be skipped.
The following table summarizes the data phase configuration in different functional modes.
Table 7. Data phase configuration versus Quad-SPI functional modes

Indirect Automatic-
Register configuration Memory-mapped mode
mode polling mode
Data read is sent back

directly over the AHB to
any master on the bus
Read data QUADSPI_DR
Data matrix requesting for
reading operation (Cortex®,
DMA LTDC, DMA2D...).
Write data QUADSPI_DR Not supported
1-byte 1-byte QUADSPI_DLR has no

0x00000000 0x00000000 meaning in this mode.
Number of data to If DMA is used, number of
QUADSPI_DLR to to
be sent/received data to be read is set only
undefined(1) 4-bytes
in DMA's DMA_SxNDTR
0xFFFFFFFF 0x00000003
register.
No Data: skipped DMODE[1:0] = 00(2)
Data on 1 line:
DMODE[1:0] = 01
Data phase Single SPI mode
QUADSPI_CCR[1 Data on 2 lines:
5:14] DMODE[1:0] = 10
Dual SPI mode
Data on 4 lines:
DMODE[1:0] = 11
Quad SPI mode
1. When QUADSPI_DLR = 0xFFFFFFFF, the number of bytes to be sent or received is undefined so the
transfer continues until the end of memory as defined in FSIZE. When QUADSPI_DLR = 0xFFFFFFFF and
FSIZE = 0x1F then the transfer continue indefinitely, stopping only after an abort request or after the Quad-
SPI is disabled. After the last memory address is read (at address 0xFFFFFFFF), the reading continues
with address = 0x00000000.
2. This mode should be used only in the Indirect mode.
22/95 AN4760 Rev 3

3.2 Multiple hardware-configurations

STM32 devices offer a very flexible Quad-SPI interface that permits the connection of
external Quad-SPI memories in different hardware configurations. The user can then
choose its own configuration.
Depending on the used hardware configuration, the number of used GPIOs can be up to 11.
The table below summarizes different use cases.
Table 8. Hardware configurations versus used GPIO number

Dual-Flash
- - Single-Flash mode
memory mode
CLK
Bank1 Bank2 BK1_IO0/SO
CLK CLK BK1_IO1/SI
Single/Dual SPI Used GPIOs BK1_IO0/SO BK2_IO0/SO BK2_IO0/SO
mode BK1_IO1/SI BK2_IO1/SI BK2_IO1/SI
BK1_nCS BK2_nCS BK1_nCS(1)
BK2_nCS(1)
GPIOs number 4 GPIOs 6 or 7 GPIOs
CLK
BK1_IO0/SO
Bank1 Bank2 BK1_IO1/SI
CLK CLK BK1_IO2
BK1_IO0/SO BK2_IO0/SO BK1_IO3
Used GPIOs BK1_IO1/SI BK2_IO1/SI BK2_IO0/SO
Quad-SPI mode
BK1_IO2 BK2_IO2 BK2_IO1/SI
BK1_IO3 BK2_IO3 BK2_IO2
BK1_nCS BK2_nCS BK2_IO3
BK1_nCS(1)
BK2_nCS(1)
GPIOs number 6 GPIOs 10 or 11 GPIOs
1. In Dual-Flash mode it is possible to use one chip select, either BK1_nCS or BK2_nCS. For more details on
dual-flash mode, refer to Section 3.2.4: Dual-Flash memory mode on page 25.
Note: If none of the phases are configured to use Quad-SPI mode, then the GPIOs corresponding
to IO2 and IO3 can be used for other functions even while QUADSPI is active.
3.2.1 Single-SPI mode (classic SPI)

This is the classic SPI where only four GPIOs are used and the data is sent on SO line and
received on SI line.
Note: Full duplex transfer is not supported.
AN4760 Rev 3 23/95

94
The IO2 and IO3 lines are optional:

• When used (IO2 and IO3 are connected to the Quad-SPI memory): IO2 and IO3 pins
should be configured as for IO0 and IO1. To allow communication with memory device:
– IO2 is in output mode and forced to ‘0’ to deactivate the “write protect” function
– IO3 is in output mode and forced to ‘1’ to deactivate the “hold” function
– This is managed by the hardware (QUADSPI peripheral) during all communication
phases
• When not used, the nWP and nHOLD memory device pins have to be connected
respectively to VDD and VSS while IO2 and IO3 pins could be used for other purposes.
In this mode, all phases as instruction, address, alternate-byte and data have to be
configured in single-SPI mode by setting the IMODE/ADMODE/ABMODE/DMODE fields in
QUADSPI_CCR to 01.
Figure 9. Hardware configuration: Single-SPI mode
Control
CLK CLK Flash communication lines
BK1_IO0/SO Q0/SI Communication line:

from QUADSPI to Flash
BK1_IO1/SI Q1/SO
QUADSPI BK1_IO2 Q2/nWP Communication line:
from Flash to QUAD-SPI
BK1_IO3 Q3/ nHOLD
Optional lines
BK1_nCS nCS
3.2.2 Dual-SPI mode

In Dual-SPI mode the hardware configuration is similar to the one in single mode, but here
two lines are used for data, it means that data is sent and received in two lines. As for
Single-SPI mode, the IO2 and IO3 lines are optional, if not used the nWP and nHOLD
device pins have to be connected respectively to VDD and VSS.
In this mode all the instruction, address, alternate-byte and data phases have to be
configured in Dual-SPI mode by setting the IMODE/ADMODE/ABMODE/DMODE fields in
QUADSPI_CCR to 10.
Figure 10. Hardware configuration: dual-SPI mode
CLK CLK Flash

Control
BK1_IO0/SO Q0/SI communication lines
BK1_IO1/SI Q1/SO
QUADSPI BK1_IO2 Q2/nWP
Communication lines
BK1_IO3 Q3/ nHOLD Optional lines
BK1_nCS nCS
MSv41113V1
24/95 AN4760 Rev 3

3.2.3 Quad-SPI mode

In the Quad-SPI mode six pins are used: four pins for data and two pins for clock and chip
select. In this hardware configuration it is possible to use either Single or Dual-SPI mode. In
Quad-SPI mode the data is transferred, received or both over four lines.
Note: In this mode, the hold and WP features are no longer available as IO3 and IO4 are used for
communications.
In Quad-SPI mode, depending on the Quad-SPI memory brand, the user can choose to
send each phase in single, dual or quad mode. Many memory manufacturers are supporting
the following configurations where Command-Address-Data: 1-1-4; 1-4-4; 4-4-4.
In general, if the address is sent in four lines then the alternate-byte should be sent in four
lines too.
Figure 11. Hardware configuration: Quad-SPI mode
CLK CLK Flash

BK1_IO0/SO Q0/SI Control
communication lines
BK1_IO1/SI Q1/SO
QUADSPI BK1_IO2 Q2/nWP
Communication lines
BK1_IO3 Q3/ nHOLD
BK1_nCS nCS
3.2.4 Dual-Flash memory mode

In Dual-Flash memory mode, the MCU communicates with two external memory devices at
the same time.This mode is useful to double throughput and to double size while using only
10 GPIOs: eight for data, one chip select for both devices and one for CLK. A throughput of
two bytes per cycle can be attained with dual-Flash memory in DDR Quad-SPI mode.
In this mode, only one chip-select could be used for both devices and then save one GPIO
for other usages and either nCS_BK1 or nCS_BK2 can be connected to both devices. The
clock has to be connected to both devices.
Different hardware configurations are allowed, offering a high flexibility to the user. Table 8
on page 23 illustrates all possible hardware configurations.
The Dual-memory mode allows doubling the throughput, as one byte can be sent or
received at every cycle. This mode is very interesting when more performance is needed;
not only throughput is doubled in Dual-Flash memory mode, but also the external memory
size is doubled.
When using two external Quad-SPI memories the size to be configured in FSIZE[4:0]
should reflect the total Flash memory capacity, which is double the size of one individual
component.
To support dual die packages with two chip-selects and dual Quad-SPI devices, the FIFO
size is always 32-bytes either in single Flash or Dual-Flash memory mode.
Note: The addressable space in Memory-mapped mode is up to 256 Mbytes either in Single-Flash
memory mode or Dual-Flash memory mode.
AN4760 Rev 3 25/95

94
The following figure shows an example of a read sequence in Dual-Flash memory Quad I/O
SDR mode.
Figure 12. Read sequence in dual-Flash memory Quad I/O SDR mode
Instruction Address Alternate Dummy Data
nCS_BK1
nCS_BK2
SCLK
BK1_IO0 4 0 20 16 12 8 4 0 4 0 4 0 4 0
Flash 1
BK1_IO1 5 1 21 17 13 9 5 1 5 1 5 1 5 1
Bytes at even
address Byte 1 Byte 3
BK1_IO2 6 2 22 18 14 10 6 2 6 2 2 2
6 6
BK1_IO3 7 3 23 19 15 11 7 3 7 3 7 3 7 3
BK2_IO0 4 0 20 16 12 8 4 0 4 0 4 0 4 0
Flash 2
BK2_IO1 5 1 21 17 13 9 5 1 5 1 5 1 5 1
Bytes at odd
Byte 2 Byte 4
address
BK2_IO2 6 2 22 18 14 10 6 2 6 2 6 2 6 2
BK2_IO3 7 3 23 19 15 11 7 3 7 3 7 3 7 3
Same command Same address Same mode bits
IO switch from
output to input
MSv41115V1
Note that all bytes at even addresses are stored in Flash 1 while all bytes at odd addresses
are stored in Flash 2. As described in Figure 12, in dual-Flash mode the same command,
address and alternate are sent to both Flash 1 and Flash 2. For example to read the first
four bytes in dual-Flash memory-mapped mode from 0x90000 000 to 0x9000 0003 the
following sequence is done by QUADSPI peripheral:
• The address 0x0000 0000 is sent to both Flashes and Byte 1 (at even address
0x9000 0000) is read from Flash 1 while Byte 2 (at odd address 0x9000 0001) is read
from Flash 2.
• Then the address 0x0000 0001 is sent to both Flashes and Byte 3 (at even address
0x9000 0002) is read from Flash 1 while Byte 2 (at odd address 0x9000 0003) is read
from Flash 2.
26/95 AN4760 Rev 3

Cautions:
• In Dual-Flash memory mode both device models must be identical, because in this
mode the same commands and addresses are issued in parallel to both Flash
memories; this permits to double the available Quad-SPI external Flash size. In the
case that the two Flash-memory devices are different, the Dual-Flash mode must be
disabled (DFM = 0) and each Flash memory could be used in standalone, allowing
either Flash 1 or Flash 2 to be enabled using QUADSPI_CR[7] FSEL bit.
• For all hardware configurations listed in the table below, each memory device is
configured in Quad-SPI mode. It is possible to connect each device in Single or Dual-
SPI mode. If DFM = 1, both devices must be configured in the same way. This permits
to double the available external data size and throughput.
• The Flash memory size, as specified in FSIZE[4:0] (QUADSPI_DCR[20:16]) should
reflect the total Flash memory capacity, which is the double of the size of one individual
component.
Table 9. Dual-Flash memory hardware configurations

Used nCS
Flash mode Hardware configuration
nCS configuration
FSEL = 0
Flash 1 CLK CLK Flash 1
enabled BK1_IO0/SO Q0/SI
Single Flash
DFM = 0(1) FSEL = 1
BK1_IO1/SI Q1/SO
BK1_IO2 Q2/nWP
Flash 2
Q3/ nHOLD
Both nCS_BK1 enabled BK1_IO3
2 nCS and nCS_BK2 BK1_nCS nCS
enabled not connected QUADSPI

together CLK Flash 2
BK2_IO0/SO Q0/SI
BK2_IO1/SI Q1/SO
Dual-Flash memory DFM
BK2_IO2 Q2/nWP
=1
BK2_IO3 Q3/ nHOLD
BK2_nCS nCS
MSv41116V1
AN4760 Rev 3 27/95

94
Table 9. Dual-Flash memory hardware configurations (continued)

Used nCS
Flash mode Hardware configuration
nCS configuration
CLK CLK Flash 1

BK1_IO0/SO Q0/SI
BK1_IO1/SI Q1/SO
BK1_IO2 Q2/nWP
BK1_IO3 Q3/ nHOLD
nCS_BK1 BK1_nCS
Dual-Flash memory DFM nCS
connected to QUADSPI
=1
both devices CLK Flash 2
BK2_IO0/SO Q0/SI
BK2_IO1/SI Q1/SO
BK2_IO2 Q2/nWP
BK2_IO3 Q3/ nHOLD
nCS
MSv41118V1
1 nCS
enabled
CLK CLK Flash 1
BK1_IO0/SO Q0/SI
BK1_IO1/SI Q1/SO
BK1_IO2 Q2/nWP
BK1_IO3 Q3/ nHOLD
nCS_BK2 nCS
Dual-Flash memory DFM
connected to QUADSPI
=1
both devices CLK Flash 2
BK2_IO0/SO Q0/SI
BK2_IO1/SI Q1/SO
BK2_IO2 Q2/nWP
BK2_IO3 Q3/ nHOLD
BK2_nCS nCS
MSv41119V1
1. When single-Flash memory mode is selected DFM = 0, the user can switch between Flash 1 or Flash 2 using FSEL bit.
Pink lines highlight the used chip select.
28/95 AN4760 Rev 3

3.2.5 DDR and SDR mode

The SDR mode is activated by default, DDR mode permits to sample at rising and falling
edge of each clock cycle and enables the possibility to double the throughput. DDR mode
allows boosting the throughput and the execution performances; it is also very useful when
the system clock (HCLK) is low and does not allow the QUADSPI to operate at maximum
speed.
• SDR: data sent on CLK falling edge and sampled on CLK rising edge.
• DDR: data sent on both CLK edges, during the address, alternate or data phases. The
sample is done one half CLK later.
When using DDR (dual data rate), also known as DTR (dual transfer rate) the user should
consider the following:
• The communication start triggering and the configuration procedure are the same as in
SDR.
• The command is sent every clock cycle like in SDR mode.
• The alternate, data and address phases are sent on both edges of the clock.
• The dummy cycles are counted every clock cycle like in SDR mode.
3.3 Three operating modes
3.3.1 Indirect mode

The Indirect mode is used in below cases:
• For reading, writing or erasing operations
• If there is no need for AHB masters to access autonomously the Quad-SPI memory
(available in Memory-mapped mode)
• For all the operations to be performed through the QUADSPI data registers using CPU
or using DMA
• To configure the Quad-SPI Flash memory.
In Indirect mode, all operations are performed through the QUADSPI register where both
read and write operations are available and managed by software. The Quad-SPI interface
behaves like a classical SPI interface. The transferred data goes through the data register
with FIFO. The data exchanges are driven by software or by DMA, using related interrupt
flags in the QUADSPI status registers.
The read and write operations are always performed in burst unless the amount of data is
equal to one. The amount of data to be transferred is set in the QUADSPI_DLR register. In
this mode it is possible to read or write data from or to external Flash memory with sizes up
to 4 Gbytes.
The Automatic-polling mode is available to generate an interrupt when the status-register
inside the Flash memory is changing (useful for checking the end of the erase or the end of
programming).
In case of an erase or programming operation, the Indirect mode has to be used and all the
operations have to be handled by software. In this case, it is recommended to use the
Status-polling mode and then poll the status register inside the Flash memory to know when
the programming or the erase operation is completed.
AN4760 Rev 3 29/95

94
3.3.2 Status-flag polling mode

The Status-flag polling mode is used in below cases:
• To read Quad-SPI Flash memory status register
• To poll autonomously for the end of an operation: QUADSPI polls the status register
inside the memory
The interface can automatically poll a specified register inside the memory and relieve the
CPU from this task (useful when polling end of programming flag for example). This is a
mode to check for example when an erase operation is completed and to know that an
interrupt could be generated.
The Quad-SPI interface can also be configured to periodically read at a defined rate a
register in the Quad-SPI Flash memory. The returned data can be masked to select the bits
to be evaluated. The selected bits are compared bit per bit with their required values stored
in the match register. The result comparison can be treated in two ways:
• ANDed mode: if all the selected bits are matching, an interrupt is generated when it
succeeds (stop on match flag)
• ORed mode: if one of the selected bits is matching, an interrupt is generated when it
succeeds (stop on match flag)
When a match occurs, the Quad-SPI interface can stop automatically. The READ STATUS
REGISTER command is used by many memory manufacturers as Micron or Spansion to
read continuously the status register.
3.3.3 Memory-mapped mode

The Memory-mapped mode is used in below cases:
• For reading operations
• To use external Quad-SPI Flash memory like an internal memory, so any AHB master
can read data autonomously
• For code execution from external Quad-SPI Flash memory.
In Memory-mapped mode the external memory is seen by the system as it was an internal
memory. This mode allows all AHB masters to access the Quad-SPI memory as an internal
memory. The CPU can execute code from the Quad-SPI memory as well.
When Memory-mapped mode is used, a prefetching mechanism fully managed by the
hardware permits the optimization of the read and the execution performances from the
external Quad-SPI memory. Given that all the communication phases such as sending
opcode or address are managed by the QUADSPI peripheral, a 32-bytes FIFO (16-bytes for
STM32L4 Series) is used for prefetching; this optimized prefetch mechanism avoids
software overhead.
The programmed instructions and frame are sent automatically when an AHB master is
accessing the memory-mapped space. Once the QUADSPI peripheral is configured, the
Quad-SPI memory is accessed as soon as there is a read request on the AHB; this is done
in the Quad-SPI memory mapped address range. This action is totally transparent for the
user.
An LTDC master for example can access autonomously to the external Flash memory
where all the access operations are fully managed by the Quad-SPI interface. Meanwhile,
the Cortex®-M CPU is executing code from the internal Flash memory.
30/95 AN4760 Rev 3

The Quad-SPI interface is able to manage up to 256 Mbytes memory starting from
0x9000 0000 to 0x9FFF FFFF in the Memory-mapped mode.
Execute in place (XIP)

The prefetch buffer supports execution in place, therefore the code can be executed directly
from the external Quad-SPI memory. The QUADSPI anticipates the next CPU access and
loads in advance the byte at the following address. If the subsequent access is indeed made
at a continuous address, the access is completed faster since the value is already
prefetched.
Figure 13. Executing non-sequential code from Quad-SPI
First read operation Jump Second read operation
nCS
Command Address Dummy Byte1 Byte2 ….. ByteN Command Address Dummy Byte1 Byte2 …..
MSv41161V1
Booting from Quad-SPI Flash memory

Boot from the Quad-SPI memory is not supported but the user can boot from the internal
Flash memory and then configure the QUADSPI in Memory-mapped mode and then the
execution starts from the Quad-SPI memory. For more details on how to execute from the
external Quad-SPI memory, refer to Section 6.2 on page 67.
Note: Reading the QUADSPI_DR in Memory-mapped mode has no meaning and returns 0.
For all the supported STM32 devices, the Quad-SPI memory is accessible by Cortex®-M
through system bus. For the STM32L4 Series, the QUADSPI is also accessible through the
I-Code and the D-Code buses when a physical remap is enabled at address 0, which allows
better execution performances.
When the QUADSPI is remapped at address 0x0000 0000, only 128 Mbytes are remapped.
Even when aliased in the boot memory space, the Quad-SPI memory is still accessible at its
original memory space.
Note: The data length register QUADSPI_DLR has no meaning in Memory-mapped mode.
AN4760 Rev 3 31/95

94
3.4 Special features
3.4.1 Send instruction only-once (SIOO)

The SIOO feature is named also by some memory manufacturers as “continuous read
mode”, “burst mode” or even as “performance enhanced mode”. This feature is available for
all Quad-SPI modes: Indirect, Automatic-polling and Memory-mapped. It is recommended to
use this feature in order to reduce command overhead and to boost the execution
performances. When SIOO is enabled, the command is sent only once when starting the
reading operation, then only the address is sent.
The command is sent only at first when starting the read operation. If a new read operation
occurs, only one address is sent; this action permits the reduction of up to eight cycles (in
Single I/O mode) for the command. This is a very interesting feature to reduce access
overhead to the Quad-SPI memory.
Data are prefetched continuously while the FIFO is not full, when a discontinuous access is
detected, the QUADSPI rises chip-select and starts a new read operation without sending
the command but sending directly the new address.
Figure 14. Executing non-sequential code from QUADSPI with SIOO enabled
First read operation Jump Second read operation Jump Third read operation
nCS
Mode .. Mode Byte Mode

Command Address Dummy Byte1 Byte2 ByteN Address Dummy Byte1 ... Address Dummy Byte1
bits . bits N bits
MSv41162V1
The SIOO feature is supported by many Quad-SPI memory manufacturers such as Micron,
Spansion and Macronix, nevertheless before using it, the user has to check if the feature is
supported by the used memory.
To enable the SIOO mode, the user should:
• Configure the memory by entering the SIOO mode. Refer to relevant manufacturer’s
datasheet for more details on how to enter this mode (make sure that the read
command to be used does support this mode). Note that an alternate byte (mode Bits)
needs to be sent in order to keep the device in this mode. Refer to SIOO example on
Section 6.2 on page 67 for more details on enabling this feature.
• Configure the QUADSPI peripheral by setting the SIOO bit in QUADSPI_CCR register.
3.4.2 Delayed data sampling

For read operations from the external memories, the delayed data sampling is useful when
the signals are delayed due to constraints on the PCB layout optimizations; hence the
optimization compensates this delay. The sampling clock can be shifted by an additional half
cycle after data is driven by the Flash memory, this is done to guarantee that the data is
ready at the sampling moment. This feature is not supported in DDR mode. To enable
sampling shift, set the SSHIFT bit in QUADSPI_CR register.
For write operations from the external memories in DDR mode, the output data can be
shifted by one quarter of the QUADSPI output clock cycle in order to relax the hold
constraints. To enable this output data delay, set the DHHC bit in QUADSPI_CR register.
32/95 AN4760 Rev 3

For more details on QUADSPI timing characteristics refer to the relevant products
datasheet.
3.4.3 Timeout counter

The timeout counter can be used to reduce the Quad-SPI memory power-consumption by
releasing the nCS and then putting the memory in a lower consumption state.
After each access in Memory-mapped mode, the QUADSPI prefetches the subsequent
bytes and holds these bytes in the FIFO. When the FIFO is full, the communication clock is
stopped but the nCS pin remains low to keep the Flash memory selected and not resend a
complete command to read the next bytes when location is available in the FIFO.
To avoid any extra power-consumption in the external Flash memory, when the clock is
stopped for a long time, the timeout counter can release the nCS pin; this action puts the
external Flash memory in a lower-consumption state after a period of timeout elapsed
without any access. Once FIFO becomes empty, the nCS is low and permits read
operations.
To use the timeout counter the user should:
• Enable it by setting the TCEN bit in the QUADSPI_CR register
• Program the timeout period TIMEOUT[15:0] in the QUADSPI_LPTR register
Note: When the timeout counter is enabled, for example in memory-mapped mode, if timeout
occurs, nCS is raised; and for any new read access, a new complete read command
sequence is started by the Quad-SPI interface.
3.4.4 Additional status bits

Other than status flags described in Table 13, the QUADSPI_SR status register includes
additional status bits, the table below summarizes the status flags.
Table 10. Additional status bits

Name Size Description
5 bits Number of valid bytes being held in the FIFO

FLEVEL
(4 bits for STM32L4 Series) (for indirect mode)
This bit is set when an operation is ongoing.
BUSY 1 bit Clears automatically when operations are
finished and FIFO is empty.
3.4.5 Busy bit and abort functionality

BUSY bit
The BUSY bit is used to indicate the state of the Quad-SPI interface, it is set in the
QUADSPI_SR register when an operation is ongoing and it clears automatically when the
operations are finished or aborted.
Note: Some QUADSPI registers cannot be written when the BUSY bit is set, so the user have to
check if it is reset before writing to registers. Refer to the relevant reference manual to check
if the register could be written or not when BUSY is set.
AN4760 Rev 3 33/95

94
The following table summarizes different cases when the BUSY bit is reset in different
QUADSPI operating modes:
Table 11. BUSY bit reset in different Quad-SPI modes

Quad-SPI mode BUSY bit reset
– The QUADSPI has completed the requested command sequence and the
Indirect mode FIFO is empty
– Due to an abort
Automatic-polling – After the last periodic access is complete, due to a match when APMS =1
mode – Due to an abort
– On a timeout event
Memory-mapped
– Due to an abort
mode
– QUADSPI peripheral is disabled
ABORT bit
When an application is running, any ongoing QUADSPI operation can be aborted by setting
the ABORT bit in the QUADSPI_CR register. Once the abort is completed, the BUSY bit and
the ABORT bit are automatically reset and the FIFO is flushed. If an abort occurs on an
ongoing AXI/AHB burst operation, the QUADSPI allows the ongoing burst to complete
properly before reseting the BUSY bit and the ABORT bit.
Note: Some Flash memories might misbehave if a write operation to a status registers is aborted.
3.4.6 4-byte address mode

This mode is named also by some brands “extended address mode”. In this mode a 4-byte
address is sent; this action permits to address the Flash memories with sizes up to four
Gbytes. This mode is supported by many Quad-SPI memory manufacturers such as Micron,
Spansion and Macronix.
Before using the 4-byte mode, the user has to check if it is supported by the used device.
To enable this mode, the user should:
• Configure the memory by entering the 4-byte mode. Refer to relevant manufacturer’s
datasheet for more details on how to enter this mode.
– For Micron devices for example, the ENTER 4-BYTE ADDRESS MODE “B7h”
command should be used, then the user should use dedicated 4-byte address
commands for some operations, such as read, program or erase, from the device
datasheet.
– For Micron devices, if a read operation is needed, the user should use 4-BYTE
READ command “13h” instead of READ “03”
• Configure the QUADSPI peripheral in 32-bits address mode by setting ADSIZE[1:0] =
11 field in QUADSPI_CCR register.
Note: The memory size has to be configured according to the fixed address size respecting the
following formula: number of bytes in Flash memory = 2[FSIZE+1] where [FSIZE+1] is
effectively the number of address bits required to address the Flash memory.
The following table summarizes the different address modes versus the maximum
addressable memory space.
34/95 AN4760 Rev 3

Table 12. Address mode versus maximum addressable memory space

QUADSPI_CCR QUADSPI_DCR Memory size
Address length
ADSIZE [1:0] [20:16] FSIZE [4:0] 2[FSIZE+1]
8-Bits address 00 00111 Up to 256 bytes

16-Bits address 01 01111 Up to 64 Kbytes
24-Bits address 10 10111 Up to 16 Mbytes
32-Bits address 11 11111 Up to 4 Gbytes(1)
1. Only the first 256 Mbytes of the Quad-SPI memory can be read in Memory-mapped mode (from
0x9000 0000 to 0x9FFF FFFF).
Cautions on 4-byte address mode:

• In Indirect mode, if the address plus the data length exceeds the Flash memory size,
TEF flag is set as soon as the access is triggered.
• In Memory-mapped mode, if an access is made to an address outside of the range
defined by FSIZE but still within the 256 Mbytes range, then an AHB error is given. The
effect of this error depends on the AHB master that attempted the access; if it is the
Cortex® CPU, a hard fault interrupt is generated and if it is a DMA, a DMA transfer error
is generated while the corresponding DMA channel is automatically disabled.
3.4.7 QUADSPI and delay block in STM32H7 Series

In the STM32H7 Series, the QUADSPI has its own delay block accessible through AHB3.
The delay block can be used to generate a sampling clock which is phase-shifted from the
output clock sent on the chip pin. The delay block aligns the sampling clock on the incoming
data.
Figure 15. QUADSPI and delay block

GPIO pin
CLK
QUADSPI
DLYB
dlyb_out_ck dlyb_in_ck
Registers access
over AHB3
dlyb_in_ck : delay block input clock
dlyb_out_ck : delay block output clock MSv61188V1
AN4760 Rev 3 35/95

94
3.5 Interrupts and DMA usage
3.5.1 Interrupts usage

The QUADSPI peripheral supports five different interrupts where each one is useful in a
particular case. To be used, each interrupt has to be enabled by setting its corresponding
enable bit and enabling the QUADSPI global interrupt on the NVIC side.
The table below summarizes all the supported interrupts.
Table 13. QUADSPI interrupts summary

Enable
Event Clear
Interrupt control Description
Flag(1) bits(3)
bit(2)
Timeout TOF TOIE CTOF Timeout occurred
Matching of the masked received data

Status match SMF SMIE CSMF with the match register (automatic-
polling mode only)
FIFO threshold FTF FTIE - FIFO threshold reached (indirect mode)
Indirect mode: the correct number of

data set in the QUADSPI_DLR register
Transfer complete TCF TCIE CTCF has been transferred
All modes: the transfer has been
aborted
Indirect mode: out-of-range address
Transfer error TEF TEIE CTEF
has been accessed
1. All event flags are available in the QUADSPI_SR register
2. All enable-control bits are available in the QUADSPI_CR register
3. All clear bits are available in the QUADSPI_FCR register
3.5.2 DMA usage

DMA can be used to perform data transfers from or to the Quad-SPI external memory, this is
possible when the Quad-SPI interface is configured either in Indirect read/write mode or in
Memory-mapped mode. In Memory-mapped mode only-read from Quad-SPI memory is
allowed.
DMA usage with QUADSPI in Indirect mode

In DMA mode, the DMA is the flow controller. When QUADSPI FIFO threshold is reached
while DMAEN bit is set, DMA requests are generated from QUADSPI to the DMA.
The transfer is started when
• The DMAEN bit is set and the QUADSPI and DMA are configured
• The FTF flag is set when FIFO threshold is attained
If DMAEN = 1 already, then the DMA controller must be disabled before changing the
FTHRES/FMODE.
36/95 AN4760 Rev 3

In indirect mode when configuring the DMA for data transfer from/to the QUADSPI, the
QUADSPI should be considered as a peripheral:
• Memory to peripheral mode in case of writing data to the QUADSPI from the internal
memory
• Peripheral to memory mode in case of reading data from the QUADSPI to be
transfered into the internal memory.
Also the address of the QUADSPI should be written into the peripheral address register
(DMA channel/stream x peripheral address).
The table below summarizes the different DMA requests and transfer directions versus the
STM32 series.
Table 14. DMA requests mapping and transfer directions versus STM32 series
Product (1) DMA1 DMA2 MDMA
Request 5 Request 3
STM32L4 Series NA
Channel 5 Channel 7
Stream 7
STM32F4 Series NA NA
Channel 3
Stream 7
STM32F7 Series NA NA
Channel 3
quadspi_ft_trg
NA NA channel
X[0..15]/Stream22
STM32H7 Series
quadspi_tc_trg
NA NA channel
X[0..15]/Stream23
1. For applicable devices of each series embedding a QUADSPI.
DMA usage with QUADSPI in Memory-mapped mode

In Memory-mapped mode the QUADSPI allows the access to the external memory for read
operation through the memory mapped address region (from 0x9000 0000 to 0x9FFF FFFF)
and allows the external memory to be seen just like an internal memory.
In that case when configuring the DMA to transfer data from the QUADSPI memory-mapped
region into an other internal memory, the QUADSPI memory-mapped region should be
considered as a memory when configuring the DMA registers:
Memory to memory mode in case of reading data from the QUADSPI memory mapped
region to be transfered into the internal memory.
Also the address of the QUADSPI should be written into the peripheral address register
(DMA channel/stream x memory address).
In Memory-mapped mode the DMA is the flow controller since the QUADSPI does not
generate DMA requests
In Memory-mapped mode either DMA1 or DMA2 can be used to transfer data from the
external Quad-SPI memory to any other memory or peripheral. To perform a transfer using
DMA from external Quad-SPI memory to any other memory or a peripheral, the user should
configure the DMA by setting the source address (from 0x9000 0000), the destination
address and the number of data to be transferred.
AN4760 Rev 3 37/95

94
The DMAEN bit has no effect in Memory-mapped mode, the transfer is started as soon as
the DMA is accessing the QUADSPI address range (from 0x9000 0000 to 0x9FFF FFFF).
Once the DMA configured transfer is started by software, the DMA reads the data from the
Quad-SPI memory exactly as an internal memory. The QUADSPI peripheral manages the
communication with the external memory and puts the read data in the FIFO.
The number of data items to be transferred is managed by the DMA so the user should
configure the number of data in the DMA’s register DMA_SxNDTR (or DMA_CNDTRx
register for STM32L4x6xx). There is no need to configure the QUADSPI_DLR register as it
has no effect in the Memory-mapped mode where the DMA is the flow controller.
Note: The DMA’s FIFO can be used for example if the DMA Burst mode is required to reduce the
transfer overhead on the bus matrix.
QUADSPI and master DMA in STM32H7 Series

In the STM32H7 Series the MDMA manages the DMA access to the Quad-SPI interface.
The MDMA can access the QUADSPI in the following ways:
• Directly from the AXI bus matrix over a 64-bit AXI bus for memory mapped access
• From AHB3 over a 32-bit AHB bus for registers access
The master DMA offers two trigger signals from the Quad-SPI interface to procure more
flexibility for the user's application. The two trigger signals are:
• quadspi_ft_trg: QUADSPI FIFO threshold trigger
• quadspi_tc_trg: QUADSPI transfer complete trigger
Figure 16. QUADSPI and master DMA
ITCM
Cortex-M7
I$ D$
DTCM
Masters accessing
QUADSPI
Slaves
MDMA
QUADSPI
quadspi_ft_trg registers access
Channel X[0..15]/Stream22
quadspi_tc_trg QUADSPI
Channel X[0..15]/Stream23 memory-mapped
region access
64-bit bus width

32-bit AHBS
64-bit AXI master bus 32-bit bus width
AXI
32-bit AHB QUADSPI registers access AHB
TCM
64-bit AXI QUADSPI memory-mapped access
MSv61185v1
38/95 AN4760 Rev 3

3.6 Low-power modes

The STM32 power state is an important requirement that must be considered as it has a
direct effect on the Quad-SPI interface state. For example, if the MCU is in Standby mode
then the QUADSPI has to be reconfigured after wakeup from this mode.
Note: The Quad-SPI memories can also be put in low-power mode. Depending on the memory
brand, some devices support both Standby mode and Deep power-down mode while other
devices support only Standby mode.
In order to save more energy when the application is in low-power mode, it is recommended
to put the Quad-SPI memory in low-power mode before entering the STM32 in low-power
mode. More information on reducing power consumption is available on Section 7.2 on
page 90.
It is possible to perform transfers in Sleep mode while the CPU is stopped thanks to the
STM32 smart architecture and to the fact that in Sleep mode all peripherals can be enabled.
This can fit wearable applications where the low-power consumption is a must.
An AHB master such as DMA could continue the transfers from the QUADSPI (when
Memory-mapped mode is used) even after entering the MCU in Sleep mode. Once the
transfer is completed an interrupt can be generated to wakeup the STM32.
Refer to the products reference manuals for low-power mode configuration details.
AN4760 Rev 3 39/95

94
QUADSPI configuration AN4760
4 QUADSPI configuration
This section describes all QUADSPI configuration steps required to perform either read,
write or erase operations.
4.1 GPIOs configuration

The user should configure the GPIOs to be used for interfacing with the Quad-SPI memory,
and this is dependent on the preferred hardware configuration. For more details on the
hardware configurations please refer toTable 8 on page 23.
Note: It is recommended to reset the QUADSPI peripheral before starting a configuration and also
to guarantee that the peripheral is in reset state.
Depending on the GPIOs availability, the user can configure either Bank1 or Bank2 GPIOs.
The user can also configure both of banks if two Quad-SPI memories are connected.
Note: All GPIOs have to be configured in high-speed mode.
4.1.1 GPIOs configuration using STM32CubeMX tool

The following example shows how to configure QUADSPI GPIOs in quad I/O mode using
Bank1 GPIOs.
Using the STM32CubeMX tool is a very simple, easy and rapid way to configure the
QUADSPI peripheral and its GPIOs as it permits the generation of a project with a
preconfigured QUADSPI.
Once the STM32CubeMX project is created, a hardware configuration can be chosen on the
Mode window. This window is found on Pinout and configuration tab under the Connectivity
section and by selecting the QUADSPI menu.
Figure 17 shows how to select the QUADSPI hardware configuration with the
STM32CubeMX where Bank1 GPIOs are used.
Figure 17. STM32CubeMX: QUADSPI GPIOs configuration
MSv61189V1
1. Dual-bank mode in STM32CubeMx refers to Dual-Flash mode.
40/95 AN4760 Rev 3

AN4760 QUADSPI configuration
If after selecting one hardware configuration (as shown in Figure 17) the used GPIOs does
not match with the memory connection board, the user can configure the alternate function
directly on the corresponding pins.
For more details on QUADSPI alternate functions availability versus GPIOs, refer to the
alternate function mapping table in the relevant datasheet.
The figure below shows how to configure manually a PF8 pin to QUADSPI_BK1_IO0
alternate function.
Figure 18. STM32CubeMX: PF8 pin configuration to QUADSPI_BK1_IO0

alternate function
The used pins are highlighted in green once the GPIOs of the Quad-SPI interface are
correctly configured.
Dual-Flash memory case

If dual bank is selected, the user should select one of the listed chip-select configurations.
For more details on different dual-Flash memory chip-select configurations refer to
Section 3.2.4: Dual-Flash memory mode.
The following figure shows how to configure chip-select 1 for both banks with
STM32CubeMX.
AN4760 Rev 3 41/95

94
Figure 19. STM32CubeMX: Dual-Flash memory QUADSPI with chip-select 1

configuration
MSv61190V1
Enabling QUADSPI interrupts

To be able to use QUADSPI interrupts, the user should enable the QUADSPI global
interrupt on the NVIC side. After that, each interrupt is enabled separately by enabling its
corresponding enable bit (interrupt-enable bits are available in the QUADSPI_CR register
described in Table 13: QUADSPI interrupts summary).
The figure below shows the configuration window where the QUADSPI global interrupt can
be enabled, under the NVIC tab.
Figure 20. STM32CubeMX: enabling QUADSPI global interrupt
MSv61191V1
42/95 AN4760 Rev 3

4.2 QUADSPI peripheral configuration and clock
4.2.1 QUADSPI peripheral configuration (QUADSPI_CR register)

The QUADSPI peripheral is configured using the QUADSPI_CR. The user must configure
the clock prescaler division factor and the sample shifting settings for the incoming data.
The DMA requests are enabled setting the DMAEN bit (DMAEN bit not available for
STM32H7 Series products). In case of interrupt usage, their respective enable bit can also
be set during this phase. The FIFO level both for DMA request generation and for interrupt
generation is programmed in the FTHRES bits.
If timeout counter is needed, the TCEN bit can be set and the timeout value can be
programmed in the QUADSPI_LPTR register.
The dual-Flash memory mode can be activated by setting DFM to 1.
The QUADSPI clock source is the AHB where a prescaler is used to generate the QUADSPI
CLK. Depending on the programmed prescaler in the QUADSPI_CR register it is possible
that both the CPU and the QUADSPI work at the same speed (PRESCALER[7:0] = 0).
The following figure shows a QUADSPI clock scheme for STM32F4, STM32L4, STM32F7,
STM32WB55xx and STM32WB15xx where QSPI_CLK = HCLK / (Prescaler + 1).
Figure 21. QUADSPI clock configuration on QUADSPI_CR register
HCLK PRESCALER[7:0] QSPI_CLK
MSv41194V1
In the STM32H7 Series devices the QUADSPI contains two different source clocks:
• quadspi_ker_ck
It is the source clock to generate QUADSPI CLK using the following relation
(QSPI_CLK=quadspi_ker_ck/(Prescaler + 1).
• quadspi_hclk (hclk3)
It is the source clock for the register interface. This clock has no impact on the
QUADSPI CLK.
STM32CubeMx permits the configuration of quadspi_ker_ck source clock in the clock
configuration section.
The following figure shows the multiple source clocks for quadspi_ker_ck using
STM32CubeMx.
AN4760 Rev 3 43/95

94
Figure 22. STM32CubeMX: quadspi_ker_ck source clock configuration in

STM32H7 Series
MSv61192V1
The source clock for quadspi_ker_ck can be selected by using the QUADSPI clock mux as
shown in the following figure.
Figure 23. STM32CubeMX: quadspi_ker_ck source clock selection in

STM32H7 Series
MSv61193V1
4.2.2 Quad-SPI Flash memory parameters configuration

(QUADSPI_DCR register)
The parameters related to the targeted external Flash memory are configured through the
QUADSPI_DCR register. The user must program the Flash memory size in the FSIZE field,
the chip-select minimum high-time in the CSHT field, and the functional mode (Mode 0 or
Mode 3) in the MODE bit in the QUADSPI_DCR register.
Note: The QUADSPI parameters can be changed when the application is running but not during
an ongoing transfer, in other words not when the busy bit is set. If a change in the
parameters is needed during an ongoing transfer, the abort bit can be used to stop the
ongoing operation, then the configuration can be changed.
44/95 AN4760 Rev 3

QUADSPI configuration using STM32CubeMX

The STM32CubeMX tool can be used to configure the QUADSPI peripheral. Once the
GPIOs have been correctly configured, as already described, the QUADSPI parameters can
be configured. However, all configurations related to starting communications have to be
added manually by the user in the project.
In the QUADSPI configuration window, select the Parameter Settings tab then configure the
parameters. The figure below shows an example of QUADSPI configuration according to
the following conditions:
• Clock prescaler = 2 => QSPI_CLK = FAHB/3
for STM32H7 Series QSPI_CLK=quadspi_ker_ck/3
• FIFO threshold = 4 => FTF flag is set as soon as there are five or more free bytes
available to be written to the FIFO in case of write operation or FTF is set if there are
five or more valid bytes that can be read from the FIFO
• Sample shifting half cycle enabled =>sample the data read from the memory half-clock
cycle later (adjust the sampling time in case of cumulated delays on PCB)
• Flash size is 16 Mbytes => number of bytes in Flash memory = 2[FSIZE+1] = 2[23+1] =>
FSIZE = 23
• Chip select high time = 2 cycles => CSHT[2:0] = 1
• Clock mode is low => clock mode 0 enabled
• Flash ID = Flash ID 1 => select the Flash memory 1 to be addressed in Single-Flash
mode FSEL=0
Note that FSEL is ignored when DFM = 1
• Dual Flash = disabled ==> Dual-Flash mode disabled DFM = 0
Figure 24. STM32CubeMX: QUADSPI peripheral configuration
QUADSPI_CR
QUADSPI_DCR
QUADSPI_CR
MSv61194V1
AN4760 Rev 3 45/95

94
4.2.3 QUADSPI and MPU configuration

The memory protection unit (MPU) can be used to make the user application more robust
and more secure by protecting a memory region.
When using an STM32 product based on a Cortex®-M7 (such as STM32F7 and
STM32H7Series) it is highly recommended to configure the QUADSPI memory region
accessible in Memory-mapped mode as strongly ordered memory.
The QUADSPI memory region accessible in Memory-mapped mode goes from
0x9000 0000 to 0x9FFF FFFF. By performing this action the CPU is prevented to access
this region while Memory-mapped mode is not enabled.
In that same area (0x9000 0000 to 0x9FFF FFFF), once the external Flash size is defined
and the Memory-mapped mode is enabled, it is recommended to configure the unused
remaining memory area also as strongly ordered memory.
For more information about configuring the memory protection unit refer to the application
note Managing memory protection unit (MPU) in STM32 MCUs (AN4838).
4.2.4 Quad-SPI memory device configuration

The Quad-SPI memory should be configured after the QUADSPI peripheral has been
configured. The Quad-SPI memory configuration depends on the selected configuration for
QUADSPI for indirect mode.
Enable writing to Quad-SPI Flash memory

The write-enable command should be sent to the memory in order to set the write-enable
latch (WEL) bit.
The WEL bit must be set prior to every programming, erasing, and write-to-the-memory
status register operation. This command is generally sent in one line without any address or
data. The used QUADSPI mode is 1-0-0 (instruction on 1 line - no address - no data).
Figure 25. Write enable sequence (command 0x06)
Configuring dummy cycle

The number of dummy cycles should be configured in the Quad-SPI memory according to
the operating clock speed, for that, the user should refer to the datasheet of the memory
device.
46/95 AN4760 Rev 3

4.2.5 Starting a communication (QUADSPI_CCR register)

Once the QUADSPI peripheral is configured, the communication can be started in one of
these modes:
• Indirect-write or Indirect-read mode
• Status-flag polling mode
• Memory-mapped mode.
The operating mode is configured in the FMODE[1:0] field in QUADSPI_CCR register.
Before starting the communication, the user should configure the frame format in the
QUADSPI_CCR register. Refer to Section 3.1: Flexible frame format for more details on
frame-format configuration.
Indirect-write mode (FMODE = 00)

Communication starts immediately if:
• A write is performed to INSTRUCTION[7:0] (QUADSPI_CCR), if no address is required
(ADMODE = 00) and no data needs to be provided by the firmware (DMODE = 00)
• A write is performed to ADDRESS[31:0] (QUADSPI_AR), if an address is necessary
(ADMODE != 00) and if no data needs to be provided by the firmware (DMODE = 00)
• A write is performed to DATA[31:0] (QUADSPI_DR), if an address is necessary (when
ADMODE != 00) and if data needs to be provided by the firmware (DMODE != 00)
Indirect-read mode (FMODE = 01)

Communication starts immediately if:
• A write is performed to INSTRUCTION [7:0] (QUADSPI_CCR), and if no address is
required (ADMODE=00)
• A write is performed to ADDRESS [31:0] (QUADSPI_AR), and if an address is
necessary (ADMODE!=00)
Status-flag polling mode (FMODE = 10)

The accesses to the Flash memory begins in the same way as in the Indirect-read mode,
communication starts immediately if:
• A write is performed to INSTRUCTION [7:0] (QUADSPI_CCR) and if no address is
required (ADMODE=00)
• A write is performed to ADDRESS [31:0] (QUADSPI_AR) and if an address is
necessary (ADMODE!=00)
Memory-mapped mode (FMODE = 11)

Once the Memory-mapped mode is configured, the communication starts as soon as there
is an access request from any AHB master.
AN4760 Rev 3 47/95

94
4.3 Hardware considerations
4.3.1 Pull-up resistance

Most Quad-SPI memory manufacturers recommend to connect a pull-up resistance to the
VCC on the CS pin. This is to ensure that, during power-up, the chip-select pin tracks its
voltage from VCC. For more details on electrical recommendations refer to the datasheet of
the relevant parts.
The following figure shows an example of a Quad-SPI memory connection where a pull-up
resistance is connected to the chip-select pin.
Figure 26. Connecting chip-select to a pull-up resistance
4.3.2 Good PCB design allows maximum QUADSPI speed

The speed at which the Quad-SPI interface operates depends on many factors, including
the board layout and the pad speeds. With a good layout it should be possible to reach the
maximum speeds described in Table 2: QUADSPI availability and features across STM32
families and committed in the datasheet of the product.
The layout should be as good as possible in order to get the best performances. To get on
PCB routing guidelines, refer to the application note Getting started with STM32F7 Series
MCU hardware development (AN4661) section “Quadrature serial parallel interface (Quad
SPI)”, available on the ST website.
4.3.3 Chip-select high time (CSHT)

When the QUADSPI sends two commands, one immediately after the other, it raises the
chip-select signal (nCS) high between the two commands for only one CLK cycle by default.
If the Flash memory requires more time between commands, the chip-select high time
CSHT[2:0] field in the QUADSPI_DCR register can be used to specify the minimum number
of QSPI_CLK cycles (up to eight) that nCS must remain high.
48/95 AN4760 Rev 3

4.3.4 CKMODE
The clock mode indicates the level that CLK takes between commands when nCS is high.
Two modes are supported when nCS is high: mode 0 where CLK stays low and mode 3
where CLK stays high.
Figure 27. Chip select high time: CSHT = two clock cycles
CSHT
nCS
SCLK CLKMODE = 0
BK1_IO0 7 6 5 4 3 2 1 0 7 6 5 4 3 2 1 0
BK1_IO1
BK1_IO2
BK1_IO3
Command n Command n+1

MSv41120V1
4.3.5 Some considerations when using QUADSPI in classical SPI mode

When using the Quad-SPI interface in classical SPI mode, the user should consider the
following equivalences:
• Mode 0 for QUADSPI is equivalent to CPOL = 0 and CPHA = 0 for classical SPI
• Mode 3 for QUADSPI is equivalent to CPOL = 1, and CPHA = 1 for classical SPI
The main difference between the two modes is the clock polarity when the bus master is in
Standby mode and not transferring any data.
• SCLK stays at logic low state with CPOL = 0, CPHA = 0
• SCLK stays at logic high state with CPOL = 1, CPHA = 1
Note: Full duplex is not supported when using the Quad-SPI interface in classical SPI mode, only
half duplex is supported.
The following figure shows a classic SPI frame example highlighting the Quad-SPI clock
modes equivalence with classical SPI.
Figure 28. QUADSPI in classical SPI mode frame example
Mode 0 SCLK
Mode 3 SCLK
nCS
SI MSB
SO MSB
MSv41196V1
AN4760 Rev 3 49/95

94
Programming Quad-SPI Flash memory AN4760
5 Programming Quad-SPI Flash memory
This section describes how to program a Quad-SPI Flash memory in the following use
cases:
• For an end application development: in this case the Quad-SPI memory is
programmed during the development of the product with static data or code to be used
in the final product. A dedicated Flash memory loader is needed in order to place the
data or the code to be used in the application. The Flash loaders provided by ST can
be used for programming if the user is using one of the ST EVAL or Discovery boards,
otherwise the user should develop its own Flash memory loader.
• On-the-fly when application is running: in this case the Quad-SPI Flash memory is
used in a final product as an external mass-storage device, this permits the application
to store data any time that it is needed.
Note: For both cases the programming principle is the same. The only difference is that in the first
case, the programming operation is performed with a tool and a Flash memory loader during
the application’s development, while in the second case the programming operation is
performed during a running application in a final product. Only the Indirect mode should be
used for programming regardless if it is a writing or an erasing operation.
Depending on the used Flash memory brand, different programming commands are
available, so it is up to the user to configure the desired command supported by the device.
The instruction, address and data phases can be sent in one, two or four lines for command
phase depending on the device brand.
The 4-byte address mode can be used to program the Quad-SPI Flashes with sizes up to
4 Gbytes.
The Automatic-polling mode can be used for waiting while the programming operation is
ongoing; when the operation is completed an interrupt can be generated.
5.1 Programming code or data for an end application

This section describes how to program either static code or data to be used in the final
application.
• Programming code for end application: the code for the application is placed in the
Quad-SPI memory to be executed by the CPU and then to be extended to the on-chip
internal memory. An example of storing code in Quad-SPI Flash memory is described
in Section 6.2: Executing from external Quad-SPI memory: extend internal memory
size.
• Programming data for end application: this is useful in graphical applications, for
example to store graphic content such as icons or images. An example of storing data
in Quad-SPI Flash memory for an end application is described in Section 6.1: Reading
data from Quad-SPI memory: graphical application.
To program the Quad-SPI Flash for an end application, either the STM32 ST-LINK utility or
the integrated development environment can be used. This operation is done using the
debug interface (SW, JTAG) through the STM32.
50/95 AN4760 Rev 3

AN4760 Programming Quad-SPI Flash memory
Figure 29. Programming Quad-SPI memory through debug interface
.BIN
.HEX
QUADSPI
Quad-SPI Debug
Flash STM32 STLink
MSv41121V2
External Flash loader: A dedicated algorithm is used to perform the programming

operation. The algorithm is loaded to the STM32 through the debug interface then executed
to perform the programming operation. The inputs for the algorithm are the binary file to be
programmed.
Only the Quad-SPI Flash loaders for the memories mounted on STM32 Evaluation and
Discovery boards are provided. For other hardware, the user has to develop its own custom
loader.
5.1.1 Programming Quad-SPI Flash memory using

the STM32 ST-LINK utility
When the used IDE is not supporting the Quad-SPI memory programming capability such
as System Workbench for STM32, the user can simply use the STM32 ST-LINK utility tool.
How to create a new Quad-SPI Flash memory loader and add it to the ST-LINK
utility
For each hardware configuration and for each Quad-SPI Flash memory brand, a dedicated
Flash memory loader should be developed. The user has to develop its own dedicated
Flash memory loader (.stldr file) if the hardware used is other than ST boards.
A project is provided in the ST-LINK utility install directory “STMicroelectronics\STM32 ST-
LINK Utility\ST-LINK Utility\ExternalLoader\N25Q256A_STM32L476G-EVAL_Cube”
allowing the user to develop an external loader for a N25Q256A Flash memory on the
STM32L476G-EVAL board. This project can be easily tailored to the user dedicated
hardware to generate the external loader.
For more details on how to develop an external Quad-SPI Flash memory loader for the
STM32 ST-LINK utility, refer to the user manual STM32 ST-LINK Utility software description
(UM0892), section “Developing custom loaders for external memory” available at
www.st.com.
Caution: The tool chain/compiler used to generate the HEX/BIN file to program the Quad-SPI
memory must be exactly the same as the one used for the application development.
The dedicated Flash-memory loader has to be added to the ST-LINK utility in order to be
able to program a Quad-SPI Flash memory.
AN4760 Rev 3 51/95

94
How to add a Quad-SPI Flash memory loader to the ST-LINK utility

To add an external Flash memory loader, go to the External Loader then click on the Add
External Loader button.
Figure 30. STM32 ST-LINK utility: adding Quad-SPI Flash memory loader
A window appears where the user should select their device. See an example below:
Figure 31. STM32 ST-LINK utility: selecting Quad-SPI Flash memory loader
Note: Only one external loader can be added, otherwise the error message in the figure below
appears. The user can remove one external loader and replace it with another one if
needed.
Figure 32. STM32 ST-LINK utility: error message
52/95 AN4760 Rev 3

How to program Quad-SPI Flash memory using the ST-LINK utility

To program the Quad-SPI Flash memory follow the steps below:
1. Connect the board using the USB cable through the ST-LINK debug.
2. In the External Loader go to the added external Flash memory loader then select
Program as shown below
Figure 33. STM32 ST-LINK utility: programming Quad-SPI Flash memory
3. The following window appears allowing the user to browse to the data file to be stored
in the Flash memory, which can be a binary file, an HEX file or Motorola S-record files
(.srec or .s19).
Figure 34. STM32 ST-LINK utility: selecting HEX file for programming
AN4760 Rev 3 53/95

94
Erasing Quad-SPI memory

To perform a mass-erase operation select Mass Erase, to erase sectors click on Sector
Erase then select the sectors to be erased as shown in the figure below.
Figure 35. STM32 ST-LINK utility: erasing sectors
54/95 AN4760 Rev 3

5.1.2 Programming Quad-SPI Flash memory using IDE

Programming Quad-SPI Flash memory using Keil
In the user project either the code, the data region or both that are to be programmed in the
Quad-SPI memory have to be specified to the linker before programming.
The following example shows how to add a Quad-SPI memory dedicated load region in a
Keil scatter file where also an execute region is created, enabling the execution of the code
from the Quad-SPI memory.
; *************************************************************
; *** Scatter-Loading Description File generated by uVision ***
; *************************************************************
LR_IROM1 0x08000000 0x00100000 { ; load region size_region

ER_IROM1 0x08000000 0x00100000 { ; load address = execution address
*.o (RESET, +First)
*(InRoot$$Sections)
.ANY (+RO)
}
RW_IRAM1 0x20000000 0x00050000 { ; RW data
.ANY (+RW +ZI)
}
}
LR_QSPI 0x90000000 0xFFFFFFF {
ER_QSPI 0x90000000 0xFFFFFFF {
*.o (.textqspi)
}
}
AN4760 Rev 3 55/95

94
How to add a Quad-SPI Flash memory loader to Keil MDK-ARM

To add the Quad-SPI Flash memory loader, go to Options for Target then in the Options
Target window select the Debug tab then click on the Settings button as shown in the figure
below.
Figure 36. Adding Quad-SPI Flash memory loader to Keil MDK-ARM project
The Quad-SPI Flash memory loader is then added in the following window.
56/95 AN4760 Rev 3

Figure 37. Adding Quad-SPI Flash memory loader to Keil MDK-ARM project
In the following window, select from the list the corresponding Flash memory loader:
Figure 38. Selecting Quad-SPI Flash memory programming algorithm
Once the corresponding Flash memory loader is added it appears in the programming
algorithm list as shown in the figure below.
AN4760 Rev 3 57/95

94
Figure 39. Quad-SPI Flash memory loader programming algorithm configuration
Once the Quad-SPI Flash memory loader is added, programming can be done by clicking
on the Load button or pressing the F8 key on the keyboard.
Note: The region to be programmed is defined by default in the external loader and can be
changed by changing the start address and the size fields.
How to proceed to program Quad-SPI Flash memory only once

During an application development, the user needs to debug its project and needs to load
the code to the internal Flash memory many times. For each load-Flash operation, the
Quad-SPI Flash memory is loaded as well, which makes the loading operation too long. In
this case, if the user already loaded data to the Quad-SPI Flash and does not need to
repeat the operation, the user can simply generate the QUADSPI data related symbol
definition file and add it to the project. The symbol definition file is a *.txt file for Keil MDK-
ARM and a *.o file for IAR EWARM; it should replace, in the project, the original source code
files (*.c or *.h) that were already programmed.
Another easier alternative consists in simply removing the already added Quad-SPI Flash
memory loader.
STM32 system workbench

The STM32 system workbench does not support a Quad-SPI external Flash-loader, the
user can use the STM32 ST-LINK utility instead as previously described.
58/95 AN4760 Rev 3

5.2 Storing and erasing data on the fly during running

application
5.2.1 Storing data

It is useful in some applications to use the external Quad-SPI Flash memory for data
storage. In that case, the Quad-SPI interface has to be configured in Indirect-write mode to
permit the on-the-fly data storage. Data to be stored can be the result of a processing such
as signal processing for audio applications; it can also be storing images captured by a
camera through the digital camera interface DCMI or any other data.
Before every programming operation an erasing operation has to be performed. Indirect-
write mode has to be used for this operation.
Note: The writing speed of the external Flash memory is slower than its reading speed. As
programming operation takes a considerable period of time, the user can use Status-flag
polling mode to poll the memory status register and once operation competed an interrupt
could be enabled.
The steps to perform a programming operation are listed below:
1. Configure the Quad-SPI interface in the QUADSPI_CR and QUADSPI_DCR registers.
2. Configure the Flash memory: enable writing to Flash, set the dummy cycles number
depending on the dock speed.
3. Configure the frame format in the QUADSPI_CCR register and start the programming
sequence.
4. After the programming sequence has finished. Configure the QUADSPI in polling mode
to check the memory status and confirm if it is ready.
The following figure shows an example of a page programming sequence where the page
size is 256 bytes.
Note: Data byte 1 is written first to the data register QUADSPI_DR then data byte 256 is the last.
When writing a 32-bit word to QUADSPI_DR register, note that the LSB byte is written first
to the FIFO then transmitted first.
Figure 40. Quad I/O page program sequence (command 0x38)
AN4760 Rev 3 59/95

94
Either CPU with interrupts or DMA can be used for programming Quad-SPI memory as
follow.
1. Programming Quad-SPI memory using Indirect-write mode
When using Indirect-write mode, all programming operations are handled by software by
writing directly to the QUADSPI_DR register. An interrupt is generated when a transfer
complete is identified or if FIFO threshold is reached.
2. Programming Quad-SPI memory using Indirect-write mode with DMA
It is generally recommended to use DMA to program the Quad-SPI memory using Indirect-
write mode since it offloads the CPU, nevertheless the final recommendation depends on
the user application. In some cases, where the amount of data to be written to the memory
is relatively small, there is no need to use DMA. Once the DMA is configured and the
programming operation has started, no intervention from the CPU is needed and the
operation ends autonomously. For more details on DMA usage, refer to Section 3.5.2: DMA
usage.
3. Usage of Status-polling mode
The user can use this mode to poll the memory status register. The figure below shows an
example of a status-register reading sequence.
Figure 41. Read status register sequence (command 0x05)
5.2.2 Erasing data

An erasing operation has to be performed before every programming operation. Indirect-
write mode have to be used for this operation as well as for programming.
The configuration required to perform an erasing operation is the same as the one required
to perform a programming operation, except that no data has to be written to the memory.
As the data phase is not needed, only the instruction and the address phases are required.
As erasing operation takes a period of time, the user can use Status-flag polling mode to poll
the memory status register. Once the operation is completed an interrupt can be enabled:
The steps to perform an erasing operation are listed below:
1. Configure the Quad-SPI interface in the QUADSPI_CR and QUADSPI_DCR registers
2. Configure the Flash memory: enable writing to Flash
3. Configure the frame format and the indirect mode in the QUADSPI_CCR register
4. Send the erasing command and address if needed (address is needed for sector
erasing)
5. Put the Quad-SPI interface in Status-polling to poll the end of the operation.
60/95 AN4760 Rev 3

Most of the Flash memory devices support a sector erasing and a full-chip erasing operation
and some of them support an additional erasing operation offering more flexibility to users
applications. Refer to the manufacturer’s datasheet for more details on the supported
erasing operations.
Note: If the used memory size is larger than 16 Mbytes, 4-bytes address mode have to be used, in
this case the user should choose the 4-byte command from the memory datasheet.
Sector-erase sequence
To erase a sector on the memory, a sector-erase command and a starting sector address
should be sent.
Example: to perform a sector-erase operation on the MICRON N25Q512A memory, the
QUADSPI_CCR register should be configured as below:
QUADSPI->CCR = 0x000025D8; /* Instruction= 0xD8; IMODE = 0x01; ADMODE=
0x01; ADSIZE = 0x02 */
QUADSPI->AR = 0x00000000; /* Address 0x00000000 is sent to erase the first
sector */
See below an example of a sector-erasing sequence:
Figure 42. Sector erase sequence
Full-chip erase sequence (bulk erase)

There is no need to send an address in order to erase the whole memory, sending a
command is enough. See below an example of a chip-erase sequence:
Figure 43. Example: full chip-erase sequence
AN4760 Rev 3 61/95

94
QUADSPI application examples AN4760
6 QUADSPI application examples
This section provides some typical QUADSPI use case examples showing how to use the
interface in Indirect-mode, Status-flag polling mode and Memory-mapped mode.
Some of these examples are provided in the STM32Cube firmware package while others
are based on other application notes also available on the ST website. Some hardware
implementation examples are provided as well at the end of this section.
This section describes the following use cases:
• Memory-mapped mode: reading data in a graphical application
• Memory-mapped mode: executing code from the Quad-SPI Flash memory
• Indirect mode: storing data on-the-fly during a running application
• Indirect mode: erasing data
• Hardware implementation example
6.1 Reading data from Quad-SPI memory: graphical application

As the graphical applications require a large amount of static data (font libraries, HMI style,
icons), the external Quad-SPI Flash memory can be fully dedicated to the static data
storage. This section provides two graphic use cases where the QUADSPI is used for data
storage:
• Frame buffer content generation from the Quad-SPI memory
• Displaying images directly from the Quad-SPI memory
6.1.1 Frame buffer content generation from Quad-SPI memory

This section provides an example where the DMA2D peripheral is reading images stored in
the external Quad-SPI Flash memory to write them on the external SDRAM. It also prepares
the frame buffer content to be read by LTDC and then displayed on a TFT-LCD display.
The example is based on a software demonstration from the STM32F7Cube firmware
package available on the ST website. This example includes one project that has been
developed for the STM32F746G-DISCO board.
The project is located on the following path: STM32Cube_FW_F7_V1.14.0\Projects\
STM32746G-Discovery\Applications\QSPI\QSPI_perfs
Use case description

This example describes how to use pictures stored on the Quad-SPI Flash memory to build
frame buffer content.
The DMA2D is used to render several animated layers on the LCD-TFT. All pictures and
icons are stored in the Quad-SPI Flash memory. The DMA2D is used to transfer pictures
from the Quad-SPI Flash memory to the SDRAM memory. During this transfer, the transfer
time is measured, then the transfer speed is calculated and displayed on the TFT-LCD.
At the same speed, the DMA2D is blending images and loading them to the SDRAM (frame
buffer) while the LTDC is updating the LCD-TFT at 60 Hz.
62/95 AN4760 Rev 3

AN4760 QUADSPI application examples
Figure 44. QUADSPI usage in a graphical application
LCD-TFT
RGB
LTDC
SDRAM
frame buffer
Chrom-ART
FMC
DMA2D
QUADSPI
16 MB Quad-SPI Flash
Chrom-ART DMA2D building frame buffer content

LTDC displaying images on the LCD
Storing images and icons to be used for the application

All images and icons that are needed for the application should be stored on the external
Quad-SPI memory. Six images and three icons are available in the project, all of them are
included in the “main images.c” file, where each one is defined as a constant in a dedicated
header file.
In this project configuration, only two images (img2 and img6 defined respectively in files
“img2.h” and “img6.h”) and three icons (icon_S, icon_T and icon_M defined respectively in
the files “icon_S.h”, “icon_T.h” and “icon_M.h”) are used and need to be stored to the Quad-
SPI Flash memory.
The icons and the images are defined as constants. In order to store these icons and
images to the Quad-SPI memory, they are placed in a dedicated section “.textqspi” as
described below:
• Img2 and img6 definition
__attribute__((section(".textqspi"))) const unsigned char img2[261120] = {}
__attribute__((section(".textqspi"))) const unsigned char img6[261120] = {}
• Icon_S, icon_T and icon_M definition:
__attribute__((section(".textqspi"))) const unsigned char icon_S[30800] =
{}
__attribute__((section(".textqspi"))) const unsigned char icon_T[30800] =
{}
__attribute__((section(".textqspi"))) const unsigned char icon_M[42000] =
{}
AN4760 Rev 3 63/95

94
As explained in Section 5.1.2: Programming Quad-SPI Flash memory using IDE, to

program the Quad-SPI memory using Keil MDK-ARM or IAR EWARM, a dedicated section
for all the data to be programmed should be created.
Following the QUADSPI section “.textqspi” in the Keil MDK-ARM scatter file:
LR_IROM2 0x90000000 0xFFFFFFF { ; load region size_region
ER_IROM2 0x90000000 0xFFFFFFF { ; load address = execution address
*.o (.textqspi)
}
}
To program data to Quad-SPI Flash, the user should first check if the Quad-SPI Flash
loader is added to the project (Keil MDK-ARM or IAR EWARM). If it is already added, the
user can simply build the project and program the memory. For more details on project
configuration refer to the readme file in the project directory.
Quad-SPI interface and memory configuration

The QUADSPI is configured in Memory-mapped mode to permit the DMA2D to read images
and icons from the QUADSPI and to write into the SDRAM through the FMC interface.
The Cortex®-M7 runs at 200 MHz while the QUADSPI speed is 100 MHz. At this clock
speed, the maximal reachable throughput is 50 Mbytes per second. The QUADSPI is
configured to operate in 1-4-4 mode. Since the embedded Quad-SPI memory does not
support the DDR mode, the SDR mode is used.
The used read command is the QUAD INPUT/OUTPUT FAST READ (0xEB) which permits
to send the address in four lines and read data in four lines while the command is sent in
one line.
The STM32F7 Series smart architecture permits an offload of the CPU. The DMA2D acts as
an AHB master and performs all transfers from the QUADSPI to the frame buffer (SDRAM)
instead of the CPU. At the same time, when the LTDC is displaying graphics, the Cortex®-
M7 can execute code from the internal Flash memory.
This demonstration shows how to interface a 16-Mbyte external Flash memory with the
STM32F7x6 device.
The Quad-SPI Flash memory is used as a non-volatile support containing all graphics
(icons, images) needed in the application.
64/95 AN4760 Rev 3

Figure 45. DMA2D reading images from Quad-SPI to build frame buffer content
DTCM
ITCM
AHBS
1 1 1
OTG HS
Ethernet
DMA2D
Arm Cortex-M7
DMA1
DMA2
LDTC
USB
MAC
GP
GP
L1-cache DTCM RAM
DMA_MEM1 ITCM RAM
DMA_MEM2
DMA_P2
AXIM AHBP
DMA_P1
AXI to muti AHB

APB1
ITCM
APB2
Masters accessing
QUADSPI
ART
Slaves
Flash
memory Quad-SPI interface
64-bit AHB
32-bit AHB bus
64-bit bus matrix
64-bit AHB bus
access
SRAM2 QUADSPI memory-
AHB1 peripheral mapped region access
Bus mutliplexer
AHB2 peripheral
LTDC fetches data from
FMC framebuffer to display it
on the LCD
Registers DMA2D transfers data
from Quad-SPI memory
Memory-mapped region to SDRAM
In parallel CPU fetches
32-bit bus matrix-S instructions from Flash
AXIM
1 Mac Ethernet , LCD-TFT and DMA2D are not available on STM32F72xxx and STM32F73xxx devices.
6.1.2 Displaying images directly from the Quad-SPI memory

As previously mentioned, all AHB masters can access the Quad-SPI memory in Memory-
mapped mode. This is a very interesting feature when the application requires to display
Quad-SPI stored graphics directly on the LCD without any CPU intervention.
This section provides an example where the LTDC peripheral reads an image stored in the
external Quad-SPI Flash memory. The example is based on a software demonstration from
the application note “Managing low-power consumption on STM32F7 Series
microcontrollers” (AN4749).
The AN4749 is provided with the X-CUBE-LPDEMO-F7 software package and is available
on the ST website. It includes one project that has been developed for the STM32F746G-
DISCO board.
Note: This application note described only the QUADSPI usage in the use case.
Software demonstration description

An image “screensaver” is located in three different regions: the internal Flash memory, the
external Quad-SPI Flash memory at the address 0x90000 000 and the SDRAM at the
address 0xC000 0000. The user must choose from which region the LTDC should display
the image.
At first run, the time is displayed on the LCD-TFT for five seconds, then a menu appears
showing a box where the user can choose the memory location from which the image is
read. If the Quad-SPI Flash memory is selected, then the LTDC reads the image directly
from the external Flash memory. See a graphical view of this use case below.
AN4760 Rev 3 65/95

94
Figure 46. LTDC reading an image directly from Quad-SPI memory

DTCM
Masters accessing
ITCM
QUADSPI
Quad-SPI interface
DMA2D
LTDC
Arm Cortex-M7 DTCM RAM
L1-cache ITCM RAM Slave
AXIM 32-bit AHB bus
ART
Flash
AXI to muti AHB LTDC reading
memory image from Quad-
SPI memory

access
Registers QUADSPI
memory-mapped
Memory-mapped region region access
Bus mutliplexer
32-bit bus matrix-S
Storing the image to be used for the application

For this application the image should be stored only once to be used in the end application
(final product).
The image to be programmed is defined in the RGB565_480x272.h file as a constant data
and is located in the RGB565_480x272_qspi dedicated section. The process to place the
image in this section with Keil MDK-ARM is described below:
const uint16_t RGB565_480x272_QSPI[]
__attribute__((section(".RGB565_480x272_qspi")));/* Keil MDK-ARM: placing
the image in .RGB565_480x272_qaspi section */
const unit16_t RGB565_480x272_QSPI[]={};
As explained in Section 5.1.2: Programming Quad-SPI Flash memory using IDE, to
program the Quad-SPI memory using Keil MDK-ARM or IAR EWARM, a dedicated section
for all the data to be programmed should be created.
Following the QUADSPI section “.RGB565_480x272_qspi” in the Keil MDK-ARM scatter
file:
LR_IROM2 0x90000000 0xFFFFFFF { ; load region size_region
ER_IROM2 0x90000000 0xFFFFFFF { ; load address = execution address
*.o (.RGB565_480x272_qspi);
}
}
To program data to the Quad-SPI Flash memory, the user should first check if the Quad-SPI
Flash loader is added to the project (Keil MDK-ARM or IAR EWARM). It if is already added,
the user can simply build the project and program the memory. For more details on the
project configuration, refer to the readme file in the project’s directory.
Quad-SPI interface and memory configuration

The QUADSPI is configured in Memory-mapped mode to allow the LTDC to read the image
directly from the Quad-SPI memory.
66/95 AN4760 Rev 3

Once the QUADSPI is configured, to display the image stored in the Quad-SPI memory, the
following API is called in the LCDConf.c file: HAL_LTDC_ConfigLayer(&hltdc_F,
&pLayerCfg, 0).
pLayerCfg is the pointer to a LTDC_LayerCfgTypeDef structure that contains the address of
the image in the Quad-SPI memory.
6.2 Executing from external Quad-SPI memory: extend internal

memory size
Using the external Quad-SPI memory permits an extension of the total available memory
space of the application. The STM32F746 device embedded on the Discovery or the
Evaluation boards integrates 1-Mbyte Flash memory; hence connecting an external Quad-
SPI Flash memory extends the available memory space to 64 Mbytes for the evaluation
board.
This section describes how to use the external Quad-SPI memory to extend the internal
Flash memory in order to enable code execution from the external Quad-SPI memory. The
software demonstration from application note STM32F7 Series system architecture and
performance (AN4667) is selected as a reference to show how to:
• Configure the Quad-SPI in Memory-mapped mode during the system initialization and
before jumping to the Quad-SPI memory code
• Place application’s code in the external Quad-SPI memory
Software demonstration description

The AN4667 is provided with the X-CUBE-32F7PERF embedded software package
available on the ST website which includes two projects: STM32F7_performances and
STM32F7_performances_DMAs. Both projects are provided with Keil MDK-ARM tool chain.
This section focuses on STM32F7_performances project.
The STM32F7_performances project shows STM32F746 device performance when
executing code from the internal and external memories. It includes seven configurations
where each one allows to select the data and the code’s locations. Since this section focus
on describing the code execution from the Quad-SPI memory, only 6_1-Quad-SPI_rwRAM-
DTCM and 6_2-Quad-SPI_rwRAM-DTCM configurations are taken into consideration for
the demonstration description.
The number of cycles consumed by the FFT process is calculated based on the system-tick
timer. The example was run on the STM32756G-EVAL and the results are shown on the
LCD-TFT or on the hyperterminal through the UART or on the IDE printf viewer.
The configuration is displayed and shows the current project configuration, the system
frequency, the different configurations of caches, ART, ART-Prefetch (ON/OFF) and the
memory configuration in case of an external memory (SDRAM or Quad-SPI).
The software demonstration is developed for STM32756G-EVAL board and can be easily
tailored to the STMF746-DISCO board. For more details on this application note refer to the
AN4667 document available at www.st.com.
The STM32f7_performances project is located in the following path: x-cube-
32F7perf\STM32CubeExpansion_AN4667_F7_V4.0.0\Projects\STM32756G_EVAL\stm32f
7_performances.
AN4760 Rev 3 67/95

94
The figure below highlights the two project configurations in Keil MDK-ARM that are
described in this document.
Figure 47. Project configurations: executing code from Quad-SPI Flash memory
Projects configuration
For both project configurations, 6_1-QuadSPI_rwRAM-DTCM and 6_2-QuadSPI_rwRAM-
DTCM, the user can change the desired QUADSPI settings in the Options for Target box as
shown in the next figure.
Note that the operating system clock during system initialization is 16 MHz, so at this
moment the QSPI_CLK = fAHB/1 = 16 MHz (by default the prescaler = 0). Once the system
initialization is done, the CPU jumps to the main function (in arm_fft_bin_example_f32.c file)
where the system clock configuration is performed. The system clock is configured to run at
216 MHz.
Both project configurations have the following QUADSPI settings:
• QSPI_CLOCK_PRESCALER = 3
• System clock is 216 MHz => QSPI_CLK = 54 MHz
• QSPI_DDRMODE => DDR mode enabled
• QSPI_INSTRUCTION_1_LINE => instruction is issued in one line
• QSPI_XIP_MODE => execute in place with SIOO enabled.
68/95 AN4760 Rev 3

Figure 48. Changing QUADSPI configuration in the project settings
6.2.1 Configuring Quad-SPI in Memory-mapped mode during system

initialization
Boot from Quad-SPI Flash memory is not supported, so the user can boot from internal
Flash memory, configure the QUADSPI peripheral in Memory-mapped mode and then jump
to execution from the external Quad-SPI memory.
In this example, the QUADSPI configuration is performed during the system initialization in
the SystemInit_ExtMemCtl() function in system_stm32f7xx.c file. All required QUADSPI
peripheral and Quad-SPI memory configurations are done in system-stm32f7xx.c file and
are described below.
AN4760 Rev 3 69/95

94
GPIOs configuration
As shown in the following figure, the Quad-SPI Flash memory is connected in Quad I/O
mode, so six GPIOs have to be configured for the Quad-SPI interface.
Figure 49. Quad-SPI Flash memory connection in STM32756-EVAL board
The Quad-SPI interface GPIOs configuration in the SystemInit_ExtmemCtl() function is

described below:
RCC->AHB1ENR |= 0x00000022; /* Enable GPIOB and GPIOF interface clock */
/* Connect PB2 and PB6 pins to Quad-SPI Alternate function */
GPIOB->AFR[0] = 0x0A000900;
GPIOB->AFR[1] = 0x00000000;
/* Configure PBx pins in Alternate function mode */
GPIOB->MODER |= 0x00002020;
/* Configure PBx pins speed to 100 MHz */
GPIOB->OSPEEDR |= 0x00003030;
/* Configure PBx pins Output type to push-pull */
GPIOB->OTYPER = 0x00000000;
/* No pull-up, pull-down for PBx pins */
GPIOB->PUPDR |= 0x00000000;
/* Connect PF6, PF7, PF8 and PF9 pins to Quad-SPI Alternate function */
GPIOF->AFR[0] |= 0x99000000;
GPIOF->AFR[1] |= 0x000000AA;
/* Configure PFx pins in Alternate function mode */
GPIOF->MODER |= 0x000AA000;
/* Configure PFx pins speed to 100 MHz */
GPIOF->OSPEEDR |= 0x000FF000;
/* Configure PFx pins Output type to push-pull */
GPIOF->OTYPER = 0x00000000;
/* No pull-up, no pull-down for PFx pins */
GPIOF->PUPDR = 0x00000000;
70/95 AN4760 Rev 3

Enabling QUADSPI peripheral

In order to configure the Quad-SPI memory, the QUADSPI peripheral should be enabled so
it can communicate with the external memory.
/* Enable the Quad-SPI interface clock */
RCC->AHB3ENR |= 0x00000002;
/* Reset QSPI peripheral */
RCC->AHB3RSTR |= (RCC_AHB3RSTR_QSPIRST); /* Reset */
RCC->AHB3RSTR &= ~(RCC_AHB3RSTR_QSPIRST); /* Release reset */
/* Enable Quad-SPI peripheral */
QUADSPI->CR = 0x00000001;
Quad-SPI memory configuration

Once the QUADSPI peripheral is enabled, it is possible to communicate with the Quad-SPI
memory in order to configure it in the desired operating mode. Referring to the MICRON
N25Q512A datasheet, the SDR single-SPI mode can be used to communicate with the
memory, meaning that the command, address and data are sent in one line in order to
configure the memory.
Note that the indirect Quad-SPI mode have to be used for external memory configuration.
Resetting the Quad-SPI memory:
Before resetting the Quad-SPI memory registers, the RESET ENABLE command 0x66
should be sent in 1-0-0 mode, so only the command is sent in Indirect mode while the
address and the data phases are skipped.
/* Send RESET ENABLE command (0x66) to allow memory registers reset*/
QUADSPI->CCR = 0x00000166;
To reset the Quad-SPI memory registers, the RESET command 0x99 should be issued in 1-
0-0 mode, only the command is sent in Indirect mode while the address and the data
phases are skipped.
/* Send RESET command (0x99) to reset the memory registers*/
Configure the memory to receive commands in four lines:
In both project configurations, the Quad-SPI memory is configured to receive commands in
one line (QSPI_INSTRUCTION_1_LINE defined), but the memory can be configured to
receive the commands in four lines. If this is the desired mode, the user should use the
QSPI_INSTRUCTION_4_LINES define instead of the QSPI_INSTRUCTION_1_LINE.
To enable the QUADSPI operation, the enhanced volatile configuration register of the
N25Q512A external memory should be configured with 0x7F.
To write 0x7F to the enhanced volatile configuration register, the 0x61 command should be
sent. In this case, the QUADSPI should send 0x61 command and 0x7F data while no
address needs to be sent, then the used mode is 1-0-1.
/* Enable write cmd : 0x06. This to allow to write to enhanced volatile
register to allow instructions to be writen in 4 lines*/
while(QUADSPI->SR & 0x20); /* Wait for busy flag to be cleared */
AN4760 Rev 3 71/95

94
/* Write to Enhanced Volatile Configuration Register of the external memory

(MT25QL512): Enable quad I/O command input. Write to enhanced volatile
configuration register cmd = 0x61, Configuration: 0x7F*/
while(!(QUADSPI->SR & 0x04)); /* Wait for FTF flag to be set */

QUADSPI->DR = 0x7F;
while(!(QUADSPI->SR & 0x02)); /* Wait for TCF flag to be set */
Enabling SIOO mode (named XIP mode in MICRON’s datasheet):
/* Enable write cmd: 0x06. This is done to allow to write to volatile
configuration register. For more details refer to MT25QL512 datasheet. */
QUADSPI->CCR = ( 0x0106 | QSPI_CCR_IMODE );

/* Configure the Quad-SPI in 1-0-1 mode to write to VOLATILE CONFIGURATION
REGISTER*/
QUADSPI->CCR = (0x00000081 | QSPI_CCR_IMODE | QSPI_CCR_DMODE );
while(!(QUADSPI->SR & 0x04)); /* Wait for FTF flag to be set */

/* Write 0x83 to volatile configuration register: bit 3 = 0 to enable XIP,
and bits [7:4] = 8 to set eight dummy cycles*/
QUADSPI->DR = ((MEM_DUMMY_CYCLE_XIP << 4) | 0x3 );
while(!(QUADSPI->SR & 0x02)); /* Wait for TCF flag to be set */
QUADSPI peripheral configuration

Configure QUADSPI peripheral (QUADSPI_CCR register):
Once the Quad-SPI memory is configured, the Quad-SPI interface should be configured in
Memory-mapped mode; the frame format is set in QUADSPI_CCR register as described
below:
• Use DTR QUAD INPUT/OUTPUT FAST READ command: INSTRUCTION [7:0] =
0xED
• Send command in one line: IMODE = 0b01
• Send address in four lines: ADMODE = 0b11
• Configure 3 bytes address: ADSIZE = 0b10
• Send alternate-byte in four lines: ABMODE = 0b11
• Configure 1 alternate-byte: ABSIZE = 0b00
• Receive data in 4 lines: DMODE = 0b11
• Configure 7 dummy cycles: DCYC = 0b00111
• Enable memory-mapped mode: FMODE = 0b11
• Enable SIOO mode: SIOO = 1
• Enable DDR mode: DDRM = 1
72/95 AN4760 Rev 3

As SIOO mode is enabled (called XIP in MICRON datasheet), an alternate-byte have to be

sent (QUADSPI_ABR = 0x00) at every new read sequence in order to keep the Quad-SPI
memory in SIOO mode.
According to MICRON’s datasheet, a latency of eight dummy cycles before receiving data
should be set when using 0xED command in Extended-SPI mode. It depends also on the
QSPI_CLK.
Only seven dummy cycles are configured for QUADSPI peripheral, however it is eight
cycles on the Quad-SPI memory side; this is due to the additional alternate-byte cycle that is
needed in order to send one byte in DDR mode.
After the address phase, there are in total a latency of eight cycles before the data phase:
seven dummy cycles + one alternate-byte cycle.
See below the configuration code:
QUADSPI -> CCR = (0x0F002C00 | QSPI_CCR_DDRM | QSPI_CCR_DCYC |
FAST_READ_CMD | QSPI_CCR_IMODE | QSPI_CCR_SIOO | QSPI_CCR_ABMODE);
Configure QUADSPI peripheral (QUADSPI_CR register)

The user can choose to operate either in DDR or in SDR mode, depending on the project
configuration. The configuration by default is the QSPI_DDRMODE defined in both
configurations. Since the DDR mode is enabled, the delayed sample shifting must be
disabled.
The following code describes how to enable or disable the DDR mode and how to configure
the prescaler and the Quad-SPI memory size:
#ifdef QSPI_DDRMODE
QUADSPI->CR |= QSPI_CLK_PRESCALER; /* SSHIFT = 0 delayed sample shifting
disabled in DDR mode */
#else
QUADSPI->CR |= QSPI_CLK_PRESCALER | 0x10 ; /* 0x10: SSHIFT = 1 */
#endif
QUADSPI->DCR = 0x00190000; /* Memory size: 512 Mb (64MB): 2^(26-1) ->
2^(25) -> 2^(0x19)*/
#endif
6.2.2 Placing application code in external Quad-SPI memory

The code to be loaded in the Quad-SPI memory consists of calculation algorithms used to
get the maximum energy bin in the frequency domain of an input signal using complex FFT,
complex magnitude, and maximum functions.
To place this code in the external Quad-SPI memory a dedicated load region should be
created in the linker file of the project. Two project configurations are available in the project:
one where the code of the application and the constant data are both placed in the Quad-
SPI memory, and the other where the code of the application is placed in the Quad-SPI
memory where the constant data is placed in the ITCM Flash. For both project
configurations the L1-DCache is enabled.
AN4760 Rev 3 73/95

94
Code and constant data are all placed in Quad-SPI memory:

6_1-Quad-SPI_rwRAM-DTCM
In this project configuration, the application code and its related constant data are both
placed in the Quad-SPI memory; so the Cortex®-M7 have to fetch them from the external
memory.
All remaining project codes as the peripheral drivers and the vector tables are placed in the
Flash memory ITCM. The following figure describes the 6_1-Quad-SPI_rwRAM-DTCM
project configuration.
Figure 50. 6_1-Quad-SPI_rwRAM-DTCM project configuration: code and data in

Quad-SPI memory
DTCM
ITCM
AHBS
OTG HS
Ethernet
DMA2D
Arm Cortex-M7
DMA1
DMA2
LDTC
USB
MAC
GP
GP
L1-cache DTCM RAM

ITCM RAM
DMA_MEM1
DMA_MEM2
DMA_P2
AXIM AHBP
AXI to muti AHB
ITCM
Masters accessing
QUADSPI
All remaining project’s
ART code + data Slaves
Quad-SPI interface
Flash
memory 32-bit AHB bus
64-bit AHB
64-bit bus matrix 64-bit AHB bus
QUADSPI registers
SRAM access
QUADSPI memory-
Quad-SPI Flash
Bus mutliplexer
Application’s code
+ constant data Code execution from
QUADSPI with L1-Cache
Registers ZI + RW + Stack + Heap

in DTCM
Memory-mapped
y pp region
g
Flash ITCM ART on
32-bit bus matrix-S
To place code and constant data in the Quad-SPI memory a dedicated load region has to be
created as shown in the following Keil MDK-ARM scatter file:
; *************************************************************
; *************************************************************
ER_IROM1 0x00200000 0x00100000 { ; load address = execution address
*.o (RESET, +First)
*(InRoot$$Sections)
; Place all remaining code and const data in Flash TCM.
.ANY (+RO)
}
}
74/95 AN4760 Rev 3


ER_IROM2 0x90000000 0x00100000 { ;load address = execution address
arm_fft_bin_example_f32.o (+RO-CODE)
arm_bitreversal2.o (+RO-CODE)
arm_cfft_f32.o (+RO-CODE)
arm_cfft_radix8_f32.o (+RO-CODE)
arm_cmplx_mag_f32.o (+RO-CODE)
arm_max_f32.o (+RO-CODE)
arm_fft_bin_example_f32.o (+RO-DATA)
arm_common_tables.o (+RO-DATA)
arm_const_structs.o (+RO-DATA)
}
RAM_RW_ZI 0x20000000 0x4000 {
.ANY (+RW +ZI)
}
RAM_STACK 0x20004000 0x4000 {
.ANY (STACK)
}
RAM_HEAP 0x20008000 0x8000 {
.ANY (HEAP)
}
}
Code placed in Quad-SPI memory while constant data in Flash memory ITCM:
6_2-Quad-SPI_rwRAM-DTCM
In this project configuration, the application code is placed in the Quad-SPI memory while its
related constant data is placed in the Flash ITCM. The Cortex®-M7 have to fetch code from
the Quad-SPI memory and data from the Flash ITCM.
All remaining project codes as the peripheral drivers and the vector tables are placed in the
Flash memory ITCM. The figure below describes the 6_2-Quad-SPI_rwRAM-DTCM project
configuration.
AN4760 Rev 3 75/95

94
Figure 51. 6_2-Quad-SPI_rwRAM-DTCM project configuration: only code in Quad-SPI

memory
DTCM
ITCM
AHBS
OTG HS
Ethernet
DMA2D
Arm Cortex-M7 DMA1
DMA2
LDTC
USB
MAC
GP
GP
L1-cache DTCM RAM
ITCM RAM
DMA_MEM1
DMA_MEM2
AXIM AHBP DMA_P2
AXI to muti AHB
* All remaining
ITCM
Masters accessing
project’s code + data
QUADSPI
* Application’s
constant data Slaves
ART
Quad-SPI interface
Flash
memory 32-bit AHB bus
64-bit AHB
64-bit bus matrix 64-bit AHB bus
QUADSPI registers
SRAM access
QUADSPI memory-
Quad-SPI Flash
Bus mutliplexer
Application’s code Code execution from
QUADSPI with L1-Cache
Registers ZI + RW + Stack + Heap

in DTCM
Memory-mapped
y pp region
g
Flash ITCM ART on
32-bit bus matrix-S
To place code and constant data in the Quad-SPI memory a dedicated load region has to be
created as shown in the following Keil MDK-ARM scatter file:
; *************************************************************
; *************************************************************
LR_IROM1 0x00200000 0x00100000 { ;load region size_region

*.o (RESET, +First)
*(InRoot$$Sections)
; Place all remained code and const data in Flash TCM.
.ANY (+RO)
}
}
LR_IROM2 0x90000000 0x00100000 { ;load region size_region
arm_fft_bin_example_f32.o (+RO-CODE)
arm_bitreversal2.o (+RO-CODE)
arm_cfft_f32.o (+RO-CODE)
arm_cfft_radix8_f32.o (+RO-CODE)
arm_cmplx_mag_f32.o (+RO-CODE)
arm_max_f32.o (+RO-CODE)
}
76/95 AN4760 Rev 3

RAM_RW_ZI 0x20000000 0x4000 {

.ANY (+RW +ZI)
}
RAM_STACK 0x20004000 0x4000 {
.ANY (STACK)
}
RAM_HEAP 0x20008000 0x8000 {
.ANY (HEAP)
}
}
AN4760 Rev 3 77/95

94
Performances analysis
The results are obtained with the STM32756G-EVAL, the CPU is running at 216 MHz,
VDD=3.3 V and with seven wait-states access to the internal Flash memory. The QUADSPI
is configured in DDR 1-4-4 mode with SIOO enabled and QSPI_CLK = 54 MHz.
The table below shows the obtained results for FFT demonstration for MDK-ARM in each
configuration.
Table 15. Execution performances versus configuration

CPU cycle
Feature configuration Memory location configuration
number(1)
- 5-RAMITCM_rwRAM-DTCM 112428
I-cache + D-cache ON (constant data in Quad-SPI memory) 6_1-Quad-SPI_rwRAM-DTCM 171056
I-cache + ART + ART-PF ON (constant data in Flash TCM) 6_2-Quad-SPI_rwRAM-DTCM 126900
1. The number of cycles may change from a version to another of the tool chain.
If the results of the case one and the case two of the “6-Quad SPI_rwRAM-DTCM”
configuration are compared, we note that there is a significant difference in terms of
performance since the demonstration uses a huge constant data.
• For the case one (6_1-Quad SPI_rwRAM-DTCM), since the read-only data and the
instructions are both located in the Quad-SPI Flash memory, a latency occurs due to
the concurrency access of the instruction fetch and the read-only data on the Quad-SPI
interface.
• For the case two (6_2-Quad SPI_rwRAM-DTCM), the read-only data and code are
separated. The read-only data is located in Flash-TCM, therefore, the concurrency of
the read-only data and the instruction fetch is avoided and the CPU can fetch the
instruction from AXI while the data is loaded from TCM at the same time. This is the
reason why the performance of the second case is clearly better than the first one.
By comparing the case 6_2-Quad-QPI_rwRAM-DTCM with the 5-RAMITCM_rwRAM-DTCM
(which gives the best performances at 112428 CPU cycles as per AN4667 document), it is
seen that it is they are close in terms of performances.
This is an example of how important it is to benefit from the STM32F7x5/F7x6 smart
architecture (in this example) in order to improve the execution performances from the
external Quad-SPI memory. For more details on how to improve the execution
performances from Quad-SPI memory, refer to Section 7.1: How to get the best
performances.
78/95 AN4760 Rev 3

6.3 Storing (programming) data on the fly during a running

application
Based on two examples from STM32Cube firmware, this section describes how to program
a Quad-SPI Flash memory on-the-fly while running an application. The programming can be
done either with a software by writing directly to the QUADSPI_DR register or by using
DMA.
The STM32Cube firmware package provides two examples of reading and programming
the Quad-SPI memory:
• QSPI_ReadWrite_DMA: using DMA
• QSPI_ReadWrite_IT: using interrupts.
These examples are provided for STM32L476G-EVAL, STM32446E-EVAL, STM32469I-
EVAL and STM32756G-EVAL boards and can be easily tailored to Discovery boards. This
section describes the STM32756G-EVAL programming examples.
Note: Since Flash memories need to be erased before writing; the user should perform an erasing
operation before programming the Flash memory. Only the region to be programmed need
to be erased; there is no need to perform a mass-chip erase operation if only one sector is
to be programmed.
6.3.1 QUADSPI indirect write: programming Quad-SPI memory using DMA

This section describes the STM32756G-EVAL example which is available on the
STM32Cube_FW_F7 in the
“Projects\STM32756G_EVAL\Examples\QSPI\QSPI_ReadWrite_DMA” directory.
This example shows how to program the Quad-SPI memory with data from the internal
SRAM. DMA2 is used to transfer the data from the internal SRAM to the Quad-SPI
interface. The data to be written “aTxBuffer” is a buffer generated on the SRAM. It contains
the string ****QSPI communication based on DMA****.
In this example, the following sequence is done in a forever loop:
1. Sector 1 of the Quad-SPI memory is erased
2. Data is written to the memory in DMA mode
3. Data is read in DMA mode to be compared with the source
– LED1 toggles each time a new comparison is good
– LED3 is on as soon as a comparison error occurs
– LED3 toggles as soon as an error is returned by HAL API.
This section focus on writing to the Quad-SPI memory. Thanks to the STM32F7x5/F7x6
smart architecture, the DMA can be used to program or to read the Quad-SPI Flash memory
and then to offload the CPU. Once the DMA is configured and the transfer is started, no
CPU intervention is needed. An interrupt can be generated once the transfer is completed.
AN4760 Rev 3 79/95

94
As described in the figure below, the DMA reads the data “aTxBuffer” from the SRAM and
writes it to the Quad-SPI memory, in the meanwhile the CPU can execute code from the
internal Flash.
Figure 52. Indirect write mode: programming Quad-SPI memory using DMA
DTCM
ITCM
AHBS
OTG HS
Ethernet
DMA2D
Arm Cortex-M7
DMA1
DMA2
LDTC
USB
MAC
GP
GP
L1-cache DTCM RAM

ITCM RAM
DMA_MEM1
DMA_MEM2
DMA_P2
AXIM AHBP
DMA_P1
AXI to muti AHB

APB1
APB2
Masters accessing
ART
QUADSPI
Flash
Quad-SPI interface
memory
64-bit AHB
64-bit bus matrix Slave
SRAM1 32-bit AHB bus
64-bit AHB bus

SRAM2
Data is transferred
AHB1 peripheral from SRAM by DMA
AHB2 peripheral to be written to
Quad-SPI Flash
FMC
QUADSPI registers
Registers access
QUADSPI
Memory-mapped region memory-mapped
region access
32-bit bus matrix-S
Bus mutliplexer
QUADSPI GPIO and DMA configuration

The GPIOs and DMA2 are configured in the HAL_QSPI_MspInit() function in the
stm32f7xx_hal_msp.c file. The HAL_QSPI_MspInit() function is called in the
HAL_QSPI_Init() function which include all the QUADSPI required configurations.
The QUADSPI global interrupt is enabled. The GPIOs configuration is done with respect to
the Quad-SPI memory connection in the STM32756G-EVAL board.
DMA2 is configured as follows:
• DMA2 configuration: Stream 7 Channel 3 is enabled
• Memory Increment enabled
• DMA2 interrupt enabled.
80/95 AN4760 Rev 3


The QUADSPI peripheral is configured in the HAL_QSPI_Init() function as described below:
• Clock prescaler = 2 => QSPI_CLK = (HCLK / 3) = 72 MHz.
• FIFO threshold FTHRES = 0x03.
• Sample shifting delay is disabled.
• Flash size is set in QUADSPI_DCR register: the memory size is 64 Mbytes = 2[FSIZE+1]
= 2 [19 + 1] so FSIZE = 0x19.
• Chip-select high time (CSHT) is set to one cycle.
• QUADSPI peripheral is enabled by setting EN bit in QUADSPI_CR register.

As previously mentioned, the write-enable command should be sent at first. This is done by
calling QSPI_WriteEnable() function.
Start programming sequence

To start the programming sequence, the QUADSPI_CCR register is configured in the
HAL_QSPI_Command() function as described below:
• Instruction: QUAD_IN_FAST_PROG_CMD (0x32)
• IMODE: one line
• ADMODE: one line
• ADSIZE: 24 bits
• DMODE: four lines
• FMODE: indirect mode
• The number of bytes to be written is set in the QUADSPI_DLR register in the
HAL_QSPI_Command() function. Number of bytes to be written = QUADSPI_DLR+1.
The programming sequence is started in HAL_QSPI_Transmit_DMA() function, so the
following configurations are done in this function:
• DMA direction configuration: DMA_MEMORY_TO_PERIPH
• The number of bytes to be transferred in S7NDTR register (S7NDTR =
QUADSPI_DLR+1)
• The DMA transfer is enabled by setting the DMAEN bit in the QUADSPI_CR register.
The address used in the QUADSPI_AR register is 0x0000 0000 as writing is performed in
the first sector. In this way the programming sequence is immediately started once DMAEN
bit is set.
Since DMA transfer to the FIFO is faster than the QUADSPI bus, the QUADSPI controls the
transfer flow by setting the FIFO threshold flag (FTF) each time there is (FTHRESH + 1) free
bytes available to be written to the FIFO. DMA transfers are initiated only when FTF flag is
set.
AN4760 Rev 3 81/95

94
6.3.2 QUADSPI indirect write: programming Quad-SPI memory using

interrupts
This example is available for all the STM32 embedding a QUADSPI interface. The
applicable products are listed in Table 2: QUADSPI availability and features across STM32
families.
This section describes the STM32756G_EVAL example which is available on the
STM32Cube_FW_F7 in the directory
Projects\STM32756G_EVAL\Examples\QSPI\QSPI_ReadWrite_IT.
This example shows how to program the Quad-SPI memory with data from the internal
SRAM using CPU and interrupts. The data to be written (aTxBuffer) is a buffer generated on
the SRAM.
Figure 53. Indirect write mode: programming Quad-SPI memory using interrupt
DTCM
ITCM
AHBS
OTG HS
Ethernet
DMA2D
DMA1
LDTC
DMA2
USB
MAC
Arm Cortex-M7
GP
GP
L1-cache DTCM RAM

ITCM RAM
DMA_MEM1
DMA_MEM2
DMA_P2
AXIM AHBP
DMA_P1
AXI to muti AHB APB1
APB2
Masters accessing
ART
QUADSPI
Flash
Quad-SPI interface
memory
64-bit AHB
64-bit bus matrix Slave
SRAM1 32-bit AHB bus
64-bit AHB bus

SRAM2
Data is transferred
AHB1 peripheral from SRAM by DMA
AHB2 peripheral to be written to
Quad-SPI Flash
FMC
QUADSPI registers
Registers access
QUADSPI
Memory-mapped region memory-mapped
region access
32-bit bus matrix-S
Bus mutliplexer
QUADSPI GPIO configuration

The GPIOs and DMA2 are configured in the HAL_QSPI_MspInit() function in the
stm32f7xx_hal_msp.c file. Note that the HAL_QSPI_MspInit() function is called in the
HAL_QSPI_Init() function which include all Quad-SPI required configurations.
The QUADSPI global interrupt is enabled. The GPIOs configuration is done with respect to
the Quad-SPI memory connection in the STM32756G-EVAL board.
82/95 AN4760 Rev 3


The QUADSPI peripheral is configured in the HAL_QSPI_Init() function as described below:
• Configure clock prescaler = 2 => QSPI_CLK = (HCLK / 3) = 72 MHz.
• FIFO threshold FTHRES = 0x03 => The Quad-SPI interrupts the CPU to write into the
FIFO. Each time FTF flag is set (each time there is (FTHRESH + 1), free bytes
available to be written to the FIFO).
• Sample shifting delay is disabled.
• Flash size is set in QUADSPI_DCR register: the memory size is 64 Mbytes = 2[FSIZE+1]
= 2 [19 + 1] so FSIZE = 0x19.
• Chip-select high time (CSHT) is set to one cycle.
• QUADSPI peripheral is enabled by setting EN bit in QUADSPI_CR register.

As previously mentioned, the write-enable command should be sent at first. This is done by
calling QSPI_WriteEnable() function.
Start programming sequence

To start the programming sequence, the QUADSPI_CCR register is configured in the
HAL_QSPI_Command() function as described below:
• Instruction: QUAD_IN_FAST_PROG_CMD (0x32)
• IMODE: one line
• ADSIZE: 24 bits
• DMODE: four lines
• FMODE: indirect mode
• The number of bytes to be written is set in the QUADSPI_DLR register in the
HAL_QSPI_Command() function.
The programming sequence is started in HAL_QSPI_Transmit_IT() function, so the
following configurations are done in this function:
• Enable the QUADSPI transfer error TEIE, FIFO threshold FTIE and transfer complete
TCIE Interrupts
Note that the used address is 0x0000 0000 in the QUADSPI_AR register as writing is
performed in the first sector.
Once that the FTF flag is set, an interrupt is generated to the CPU. The CPU jumps from the
main code to the interrupt routine HAL_QSPI_IRQHandler() located in
stm32f7xx_hal_qspi.c and checks the source of the interrupt, then the CPU starts
transferring data to the FIFO. Once finished, the CPU clears the FTF flag, exits the
HAL_QSPI_IRQHandler() and returns to the main code.
AN4760 Rev 3 83/95

94
6.4 Erasing-data example

This section describes how to erase the Quad-SPI Flash memory. As previously mentioned,
the Indirect mode should be used for Quad-SPI Flash memory erasing.
All the provided Quad-SPI examples in the STM32Cube firmware package include an
erasing operation example. For the STM32F7x5/F7x6 products EVAL board, the examples
are available in the following STM32Cube directory:
STM32Cube_FW_F7_VX.X.X\Projects\STM32756G_EVAL\Examples\QSPI.

Before performing an erasing operation, a write-enable command has to be sent to the
memory, this is done using the QSPI_WriteEnable() function.
Start erasing sequence

To start erasing sequence, the QUADSPI_CCR register is configured in the
HAL_QSPI_Command_IT() function as described below:
• Instruction: SECTOR_ERASE_CMD (0xD8)
• IMODE: one line
• ADSIZE: 24 bits
• DMODE: no data
• FMODE: indirect write mode
• QUADSPI_AR = 0x0000 0000.
After configuring the QUADSPI_CCR register, and since no data need to be sent, the
erasing sequence is started immediately once the address of the first sector is provided.
84/95 AN4760 Rev 3

6.5 Hardware implementation example

This section provides some hardware implementation examples of connecting the Quad-
SPI Flash memory to the STM32 based on the existing ST Discovery and Evaluation
boards. The following table summarizes the different STM32 boards embedding Quad-SPI
Flash memory.
Table 16. Different STM32 boards embedding Quad-SPI Flash memory

Product families Board Quad-SPI Flash model Size (Mbytes)
STM32L475 B-L475E-IOT01A MX25R6435F 8

STM32L476G-DISCO N25Q128A13EF840E 16
STM32L476
STM32L476G-EVAL N25Q256A13EF840E 32
STM32L496 STM32L496G-DISCO MX25R6435FM2IL0 8
STM32F412 STM32F412G-DISCO N25Q128A13EF840F 16
STM32F413 STM32F413H-DISCO - 16
STM32F446 STM32446E-EVAL N25Q256A13EF840E 32
STM32F469I-DISCO N25Q128A13EF840F 16
STM32F469
STM32469I-EVAL MT25QL512ABA8ESF-0SIT 64
STM32F479 STM32479I-EVAL MT25QL512ABA8ESF-0SIT 64
STM32F723 STM32F723e-DISCO MX25L51245G 64
STM32F746G-DISCO N25Q128A13EF840E 16
STM32F746
STM32746G-EVAL MT25QL512ABA8ESF-0SIT 64
STM32F750 STM32F7508-DISCO - 16
STM32F756 STM32756G-EVAL N25Q512A13GSF40E 64
STM32F769I-EVAL MT25QL512ABA8ESF-0SIT 64
STM32F769 MT25QL512ABB1EW9
STM32F769I-DISCO 64
/MX25L51245G
STM32F779 STM32F779I-EVAL N25Q512A13GSF40E 64
STM32H743/STM32H753 STM32H7xxI-EVAL MT25TL 01GHBB8ESF-0SIT 128
AN4760 Rev 3 85/95

94
STM32F746G-DISCO discovery board

The figure below shows an example of a connected MICRON Quad-SPI Flash memory in
Quad I/O mode on the STM32F746G-DISCO discovery board.
Figure 54. Quad-SPI memory connection on the STM32F746G-DISCO discovery board
STM32L476G-EVAL board
The figure below shows an example of how to connect MICRON Quad-SPI Flash memory in
Quad I/O mode on the STM32L476G-EVAL discovery board.
Figure 55. Quad-SPI memory connection on the STM32L476G-EVAL board
86/95 AN4760 Rev 3

AN4760 Performance and power
7 Performance and power
This section presents some recommendations on how to improve performance and how to
decrease power consumption for applications using the Quad-SPI interface.
7.1 How to get the best performances
7.1.1 Write performance

Since the writing speed of the Quad-SPI Flash memories is considerably low, there is limited
benefit when optimizing the transmission speed. It is beneficial to use burst (page
programming) writing to reduce the command overhead. In addition, DMA can be used to
offload the CPU.
7.1.2 Read performance

This section describes how to get the optimum reading performances using QUADSPI
peripheral features by following the described recommendations.
Configure QUADSPI at maximum speed

One of the most important parameters that permits to boost the read performances is the
QUADSPI clock that should be as high as possible. As previously mentioned (see
Section 4.3.2: Good PCB design allows maximum QUADSPI speed), the maximum
reachable QUADSPI operating speed depends mainly on the PCB design quality so the
user should optimize the PCB design.
For instance, if the user hardware allows the QUADSPI to operate at 60 MHz in DDR
Quad I/O mode, then for a sequential access case, an image can be read at
60000000/1024/1024= 57.22 Mbyte/s. In this case for a 4 Kbytes image, the total transfer
time is of 4107 cycles: eight cycles for command + three cycles for the address (24 bits in
DDR mode) + 4096 cycles for the 4 Kbytes image.
Another important parameter to be considered by the user when selecting the Quad-SPI
memory device, is the maximal clock frequency supported by the Quad-SPI memory in SDR
and DDR modes.
Use DDR mode (DTR)

Use the DDR mode to double the throughput, for instance in SDR Quad I/O mode one byte
is read every two clock cycles while in DDR mode one byte is read every clock cycle.
Another interesting usage for the DDR mode is that it can be a very good alternative in low-
power applications requiring to operate at a low system clock in order to save power where
it is not possible that the QUADSPI operates at its maximum speed. In that case the user
can use the DDR mode to double reading speed but at the same operating QUADSPI clock.
Note: The user should make sure that the read command that is used does support the DDR
mode.
AN4760 Rev 3 87/95

94
Performance and power AN4760
Use Quad I/O mode

Using Quad I/O mode permits a boost in reading performances, so the user should use the
Quad I/O mode rather than the Single or Dual I/O mode. It is recommended to use the
Quad I/O mode for all the phases: command, address, alternate-byte and data.
Note that sending the command in four lines is supported by some memory devices such as
Micron or Spansion.
Use Dual-Flash mode

Using Dual-Flash mode requires adding only four additional GPIOs (IO4 ... IO7) and
permits to double the throughput. For example in SDR Quad I/O dual-Flash mode one byte
is read each QSPI_CLK cycle, while in DDR Quad I/O dual-Flash mode two bytes are read
each QSPI_CLK cycle.
Reduce command overhead

Each access to Quad-SPI memory needs a command and an address to be sent which
leads to command overhead. In order to reduce command overhead and boost the read
performance, the user should use the following recommendations:
• Use large burst transfers for Indirect mode
Since each access to Quad-SPI memory needs to send a command and an address, it is
beneficial to perform large burst transfers rather than small repetitive transfers; this action
permits reduction of command overhead.
• Sequential access in Memory-mapped mode
The best read performance is achieved if the stored data is read out sequentially, which
avoids command and address overhead and then leads to reach the maximum
performances at the operating QUADSPI clock speed.
In general it is the case where Quad-SPI memory is used for storage applications such as
graphics or multimedia contents (see examples in Section 6.1: Reading data from Quad-SPI
memory: graphical application.)
• Consider timeout counter
The user should consider that enabling timeout counter in Memory-mapped mode may
increase command overhead. When timeout occurs, the QUADSPI rises chip-select. After
that, to read from the Quad-SPI memory a new read sequence needs to be initiated; it
means that the read command should be issued again, which leads to command overhead
(see Section 3.4.3: Timeout counter).
Timeout counter permits decreasing power consumption (see Section 7.2.1: Use timeout
counter), but if the performance is a concern, the user can increase the timeout period in the
QUADSPI_LPTR register or even disable it.
If the power consumption is also a concern, the user can enable the SIOO feature without
the need to disable the timeout counter.
88/95 AN4760 Rev 3

• Use SIOO feature (Continuous read mode) for random and non-sequential accesses
For random and non-sequential accesses, the command overhead increases. As described
in Figure 13, a command and an address are sent to the memory every new read sequence.
In this case, the user should enable the SIOO feature in order to reduce the command
overhead (see Section 3.4.1: Send instruction only-once (SIOO)).
Note: Not all the read commands support the Continuous read mode (Enhance performance
mode) so the user should consider this information when selecting the read command.
Execution performance
To improve the execution performance from the Quad-SPI Flash memory, the user should
follow the previously described read performance recommendations.
Executing from the Quad-SPI memory is generally characterized by its random and non-
sequential accesses. As already mentioned, an important recommendation to boost
execution performance is enabling the SIOO feature.
As seen in Figure 50: 6_1-Quad-SPI_rwRAM-DTCM project configuration: code and data in
Quad-SPI memory, placing both the code and the read-only data in the Quad-SPI memory
leads to concurrency on the Quad-SPI interface during execution.
In order to avoid this concurrency, the user can separate the read-only data and the code.
For example, the read-only data can be located in the Quad-SPI memory while the code can
be located in the Flash-TCM. This action permits to avoid the concurrency of the read-only
data and the instruction fetch; therefore, the CPU can fetch the instructions from Flash-TCM
while the data is loaded from the Quad-SPI memory at the same time.
When the application contains huge constants data, the user can separate constants and
code, each one in a dedicated section. If the code section size fit the internal Flash memory
size, the code can be loaded in the internal Flash memory while the constants are loaded in
the Quad-SPI memory.
For the STM32F7x5/F7x6, it is recommended to enable the Cortex®-M7 L1-Cache.
General recommendations
• Use DMA for data transfers in order to offload the CPU.
• Use Flag-status polling mode rather than software flag checking.
• Use Memory-mapped mode to permit any AHB master to access the Quad-SPI
memory without CPU intervention.
AN4760 Rev 3 89/95

94
Performance and power AN4760
7.2 Decreasing power consumption

One of the most important requirements in wearable and mobile applications is the power
consumption. Some recommendations can be followed in order to decrease the power
consumption.
To decrease the total application’s power-consumption, the user usually puts the STM32 in
low-power mode. To reduce even more the current consumption, the connected Quad-SPI
memory also can be put in low-power mode (also known as deep power-down mode).
For most Quad-SPI Flash memory devices the default mode after the powering-up
sequence is the standby mode. In standby mode, there is no ongoing operation; the nCS is
high but current consumption can be reduced even more. The DPD mode allows reducing
the power consumption.
7.2.1 Use timeout counter

The timeout nCS feature can be used to avoid any extra power-consumption in the external
Flash memory. When the clock is stopped for a long time, the timeout counter can release
the nCS pin to put the external Flash memory in a lower-consumption state after a period of
timeout elapsed without any access (see Section 3.4.3: Timeout counter).
7.2.2 Put the Quad-SPI memory in Deep power-down mode

The Deep power-down mode requires to enter the deep-power instruction. During the Deep
power-down mode, the device is not active and all of the write, program and erase
instructions are ignored. When CS# goes high, it is only in Standby mode and not in Deep
power-down mode. Deep power-down mode is different from standby mode.
Before entering a low-power mode or when the Quad-SPI memory is not used, it can be put
in DPD mode in order to reduce the overall application’s power-consumption. For several
memory brands the command 0xB9 should be sent to the external serial-Flash memory to
enter the deep power-down mode. The following figure shows the DPD sequence:
Figure 56. Deep power-down (DPD) sequence (command B9)
90/95 AN4760 Rev 3

To exit DPD mode, the RELEASE FROM DEEP POWERDOWN command (0xAB) should
be sent. The figure below shows the RPD sequence:
Figure 57. Release from deep power-down (RDP) sequence (command AB)
Note: Not all the serial Flash memories support the Deep power-down mode. If the selected
external serial memory does not support the Deep power-down mode, the STM32 may
control an external-power switch through a GPIO to remove the power supply of the external
Quad-SPI Flash memory and to cancel its current consumption.
7.2.3 Quad-SPI Flash memories supporting DPD mode

Many Quad-SPI memory brands support the DPD mode, here below an example:
• WINBOND: W25Q64CV
• Spansion: S25FL032P
• MICRON: MT25QL512AB
• Macronix: MX25L12865F
AN4760 Rev 3 91/95

94
Supported devices AN4760
8 Supported devices
The STM32 Quad-SPI interface has a very flexible frame format that permits the following:
• Send up to five phases: instruction – address – alternate byte – dummy – data
• Skip any phase
• Send each phase in one, two or four lines
• Send address in one, two, three or four bytes
• Send one, two, three or four alternate-byte
• Send up to 31 dummy clock cycles.
In addition, STM32 Quad-SPI interface permits sending any command, so the user can
program the desired command in the QUADSPI_CCR register in the INSTRUCTION[7:0]
field.
The STM32 Quad-SPI interface is fully configurable in terms of frame format and hardware
and it supports most Quad-SPI memory in the market.
There are several suppliers of QUADSPI compatible memories, such as Winbond,
Spansion, Macronix, MICRON (Numonyx), Microchip (SST) and others.
92/95 AN4760 Rev 3

AN4760 Conclusion
9 Conclusion
The STM32 devices provide a very flexible and useful Quad-SPI interface, which fits
memory hungry applications at a lower development cost. The QUADSPI avoids the
complexity of design with external parallel Flash memories by reducing the pin count and
offering better performances. This application note demonstrates the STM32 Quad-SPI
interface performances and flexibility, which allows lower development costs and faster time
to market.
AN4760 Rev 3 93/95

94
Revision history AN4760
10 Revision history
Table 17. Document revision history

Date Revision Changes
01-Apr-2016 1 Initial release.

Document updated to enlarge scope to other product families. All
02-May-2019 2
sections are impacted by the update.
Updated:
– Document title: Quad-SPI interface on STM32 microcontrollers
and microprocessors
– Table 1: Applicable products
– Section 1: General information
– Section 2.1: QUADSPI availability and features across STM32
families
– Table 3: Benefits of using STM32 Quad-SPI interface
– Figure 1: System architecture: STM32L4 Series
– Figure 2: System architecture: STM32F4 Series
– Figure 3: System architecture: STM32F7 Series
– Figure 4: System architecture: STM32H7 Series
– Section 2.3.5: System architecture: STM32WB35xx and
28-Apr-2020 3 STM32WB55xx devices
– Figure 5: System architecture:STM32WB35xx and
STM32WB55xx
– Figure 45: DMA2D reading images from Quad-SPI to build
frame buffer content
– Figure 46: LTDC reading an image directly from Quad-SPI
memory
– Figure 50: 6_1-Quad-SPI_rwRAM-DTCM project configuration:
code and data in Quad-SPI memory
– Figure 51: 6_2-Quad-SPI_rwRAM-DTCM project configuration:
only code in Quad-SPI memory
– Figure 52: Indirect write mode: programming Quad-SPI memory
using DMA
– Figure 53: Indirect write mode: programming Quad-SPI memory
using interrupt
94/95 AN4760 Rev 3

AN4760
IMPORTANT NOTICE – PLEASE READ CAREFULLY
STMicroelectronics NV and its subsidiaries (“ST”) reserve the right to make changes, corrections, enhancements, modifications, and
improvements to ST products and/or to this document at any time without notice. Purchasers should obtain the latest relevant information on
ST products before placing orders. ST products are sold pursuant to ST’s terms and conditions of sale in place at the time of order
acknowledgement.
Purchasers are solely responsible for the choice, selection, and use of ST products and ST assumes no liability for application assistance or
the design of Purchasers’ products.
No license, express or implied, to any intellectual property right is granted by ST herein.
Resale of ST products with provisions different from the information set forth herein shall void any warranty granted by ST for such product.
ST and the ST logo are trademarks of ST. All other product or service names are the property of their respective owners.
Information in this document supersedes and replaces information previously supplied in any prior versions of this document.
© 2020 STMicroelectronics – All rights reserved
AN4760 Rev 3 95/95

95

QuadSPI STM

Uploaded by

Copyright:

Available Formats

QuadSPI STM

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

QuadSPI STM

Uploaded by

Copyright:

Available Formats

AN4760

Table 1. Applicable products

STM32F7 Series, STM32L4 Series

April 2020 AN4760 Rev 3 1/95

3 Quad-SPI interface description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2/95 AN4760 Rev 3

3.4.4 Additional status bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

5 Programming Quad-SPI Flash memory . . . . . . . . . . . . . . . . . . . . . . . . 50

6 QUADSPI application examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

AN4760 Rev 3 3/95

6.1.1 Frame buffer content generation from Quad-SPI memory . . . . . . . . . . . 62

7 Performance and power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4/95 AN4760 Rev 3

Table 1. Applicable products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

AN4760 Rev 3 5/95

Figure 1. System architecture: STM32L4 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

6/95 AN4760 Rev 3

AN4760 Rev 3 7/95

This document applies to STM32 Arm®-based(a) microcontrollers and microprocessors.

2.1 QUADSPI availability and features across STM32 families

8/95 AN4760 Rev 3

Table 2. QUADSPI availability and features across STM32 families

STM32MP1 166 90 Yes

AN4760 Rev 3 9/95

2.2 Quad-SPI benefits against classic SPI and parallel interfaces

2.2.1 Main benefits of STM32 embedded Quad-SPI interface

Table 3. Benefits of using STM32 Quad-SPI interface

Supports single, dual and Quad-SPI memories.

10/95 AN4760 Rev 3

Table 3. Benefits of using STM32 Quad-SPI interface (continued)

Throughput is multiplied by four versus traditional SPI.

2.3 QUADSPI in a smart architecture

AN4760 Rev 3 11/95

2.3.1 System architecture: STM32L4 Series

Figure 1. System architecture: STM32L4 Series

Cortex-M4 DMA1 DMA2 DMA2D

SRAM2 QUADSPI registers

AHB1 peripheral QUADSPI memory-

AHB2 peripheral Bus mutliplexer

12/95 AN4760 Rev 3

2.3.2 System architecture: STM32F4 Series

Figure 2. System architecture: STM32F4 Series

CCM data Cortex-M4 ART

Masters accessing memory

Quad-SPI interface SRAM1

Bus matrix-S Memory-mapped region

AN4760 Rev 3 13/95

2.3.3 System architecture: STM32F7 Series

Figure 3. System architecture: STM32F7 Series

Arm Cortex -M7

L1-cache DTCM RAM

AXI to muti AHB

14/95 AN4760 Rev 3

2.3.4 System architecture: STM32H7 Series

Figure 4. System architecture: STM32H7 Series

AXI SRAM AHB1

Memory-mapped region APB1

64-bit AXI bus matrix

32-bit AHB bus matrix