SPI stands for Serial Peripheral Interface, and was initially developed by Motorola. It is full-duplex (data can be sent in both directions at once), and is ideally suited to sending data streams between devices. Speeds of 10MHz or more are achievable. It is a de-facto standard, which means there is no governing body that defines and regulates the protocol. This means there a quite a number of protocol variants.
- SPI has much higher throughput compared to other board-level communication protocols (such as I2C or 1-wire), primarily because the bus lines are driven both high and low, and there is a separate wire for transmit and receive (full-duplex)
- SPI can have an arbitrary data/frame length
- Most logic analysers support SPI decoding
- When configured in the standard manner, SPI requires 1 extra control line (for the slave select signal) from the master for every extra slave added to the SPI bus. This can take up more space and I/O pins for designs with a large number of SPI devices.
SPI can either be three wire (when there is only one slave and the slave does not require any signal on this line), or four wire (when there are multiple slaves, and the slave select line needs to be used). For every slave there needs to be new select line, but the other three traces can be shared.
One limitation with SPI is that the master has to initiate all communication. This can be a problem if the slave has data for the master but the master hasn’t or doesn’t know when to ask for it. Designers get around this by also providing a Data Ready line to the master. This is separate from the SPI interface, and usually set to trigger an interrupt to tell the master to request for the data.
No specific termination is needed on SPI connections. Long connections (many metres or more) and high data rates (>10Mhz) may require standard termination procedures to prevent reflections.
Many microcontrollers support the SPI protocol with dedicated hardware to perform the low-level functions associated with sending and receiving SPI data. However, SPI can also be bit banged (see below).
The MSB (most significant bit) is sent first, and naturally is then the first to be received. There is no pre-defined packet format, so there is no overhead. This makes SPI great for fast transmission of data streams.
Chip select normally uses inverse logic (low = chip selected). It usually is used to ‘frame’ a command sequence.
Because the master has to always drive the clock signal, if the slave wants to send data back to the master, the master must know about this and must have some way of knowing how many clock signals to send.
SPI has four standard ‘modes’. These define different polarities of the clock cycle and whether sampling is on the positive or falling edge of the clock. This is sometimes called the SPI Clock Polarity (CPOL) and SPI Clock Phase (CPHA).
The clock polarity (CPOL) determines whether the idle state of the clock signal is either 0 (CPOL = 0) or 1 (CPOL = 1).
The clock phase (CPHA) determines whether data is captured/sent on the rising or falling edge. If CPHA = 0, then data is sampled on the first clock edge. If CPHA = 1, then data is sampled on the second clock edge. This is true no matter what the clock polarity (CPOL) is set to. Note that if CPHA = 0, then data must be setup before the first clock edge.
The following table shows the naming conventions for Microchip PIC or ARM-based microcontrollers:
|SPI Mode||Clock Polarity (CPOL)||Clock Phase (CPHA)||Which Clock Edge Is Used To Sample/Shift?|
|0||0||0||Data sampled on rising edge and shifted out on falling edge.|
|1||0||1||Data sampled on falling edge and shifted out on rising edge.|
|2||1||0||Data sampled on falling edge and shifted out on rising edge.|
|3||1||1||Data sampled on rising edge and shifted out on falling edge.|
A common point of confusion is what clock phase (CPHA) means for data sampling/shifting for the different clock polarities. I have seen many sites and diagrams online which state that a clock phase of
0 means that data is sampled on the rising edge for
The standard defines these different modes to allow for greater variability in the master and slave devices that can use SPI.
What Is The Idle State?
The idle state is defined as the periods when:
- CS is high and transitioning to low at the start of a transmission
- CS is low and transitioning to high at the end of a transmission
Can A Single Master Support Multiple SPI Modes On The Same Bus?
The short answer is yes, as long as the master can be configured to all the relevant modes (most SPI peripherals inside microcontrollers support multiple SPI modes). The SPI slaves do not care what happens on the SCLK, MOSI and MISO lines while their chip select is inactive (high). So other slave devices that use other SPI modes can be communicated with whilst the chip select is held high for all other slave devices (as per normal operation). Care must be taken to change the clock polarity to what the slave node expects before making it’s chip select active.
SPI is inherently synchronous (requires a clock signal). There is no such thing as asynchronous SPI, as there is with UART and other transmission protocols.
Bit banging (on the master device) can be easily done with SPI since it is synchronous and the master has full control over the clock, hence the timing can be manipulated. Care has to be taken to assert the right lines and read data before applying the next clock transition, as well as obeying any minimum/maximum time specifications for each state.
SPI daisy-chaining is way of overcoming the routing/capability issue of having many SPI slaves and therefore many slave select lines. ICs have to support SPI daisy-chaining with a DOUT (or similar) signal before you can implement it.
The basic idea is that instead of data line being connected to every slave, the master’s data line is connected to one slave only, and that slaves DOUT is connected to the next slaves data in, forming a “daisy chain”. The other difference is that a single chip select line is routed to all slave ICs.
The data is passed from the microcontroller to the first slave, who stores it in a shift register. After a number of clock cycles, the data reaches the end of it’s internal shift-register, and is passed onto the next slave. The microcontroller continues passing out data until all the slaves shift-registers are full, at which point a pulse is sent down the global chip select signal, which causes the slave devices to read/execute/do whatever with the data currently in it’s shift register.
Some devices that support daisy chaining are Microchips MCP42xxx digital potentiometers and Linear Technologies LED drivers.
Some slave devices only support point-to-point SPI communication. This means that there can only be one master on the bus, and also only one slave (the device which supports point-to-point SPI).
The Freescale FXOS8700CQ magnetometer is one such example.
Dedicated Chip Select Pins
Some microcontrollers have dedicated chip select pins which are connected to the SPI peripheral inside the microcontroller. This pin usually has a number of different purposes:
- Used to select a slave device for communication (only really works when the SPI bus has only 1 slave on it)
- Synchronize data frames
- Detect conflicts between multiple masters
On STM32 microcontrollers this pin is called
NSS (which stands for not slave select, not being because the signal is active low).
The Microwire protocol is a subset of the SPI communication protocol, and is a trademark of National Semiconductor.
The mSPI bus is a modification of the SPI bus that enforces that the comms protocol always has 4-wires, no matter how many slave devices are attached. This simplifies the software needed on the slave devices. All devices share the same SS (slave select) line.
The RapidS term is used by Atmel and Adesto Technologies. It is commonly present on memory chips such as EEPROM and Flash memory ICs. The RapidS serial interface is SPI compatible for frequencies up to 33MHz. The RapidS protocol is different to the Rapid8 protocol, which is a parallel interface.