Avalon Protocol

Discover Intel/Altera's Avalon protocol for memory-mapped and streaming interfaces in FPGAs

What is Avalon?

Avalon is a family of interface protocols developed by Intel (formerly Altera) for their Cyclone, Arria, and Stratix FPGAs. It is used in Platform Designer (formerly Qsys) to connect IPs within an FPGA SoC.

The Avalon family includes several variants:

Avalon-MM (Memory Mapped): register and memory access
Avalon-ST (Streaming): continuous data stream
Avalon-Interrupt: interrupt signals
Avalon-Clock: clock distribution
Avalon-Conduit: custom out-of-band signals

Avalon-MM (Memory Mapped)

Main Signals

Signal	Direction (Master)	Description
address	Output	Word address
read	Output	Read request
write	Output	Write request
readdata	Input	Read data (slave → master)
writedata	Output	Write data (master → slave)
byteenable	Output	Active byte mask
waitrequest	Input	Slave requests a wait cycle
readdatavalid	Input	(pipeline) Read data is valid

Avalon-MM Read and Write Cycles

Avalon-MM timing diagrams

The master must hold read and address stable while waitrequest is high.

VHDL Implementation

A zero-latency Avalon-MM slave is the simplest implementation: waitrequest is always low, and reads/writes execute in a single cycle. The slave decodes the address to access an internal register array.

Avalon concepts are practiced through the protocol quizzes and exercises, including Avalon-MM slave implementation work.

Avalon-ST (Streaming)

Avalon-ST is designed for continuous data streams (audio, video, network). It uses a valid/ready handshake similar to AXI-Stream.

Main Signals

Signal	Direction (Source)	Description
data	Output	Stream data
valid	Output	Data is valid
ready	Input	Sink can accept data
startofpacket	Output	First beat of a packet
endofpacket	Output	Last beat of a packet
empty	Output	Number of empty bytes in the last beat
error	Output	Error associated with the beat

valid/ready Handshake

A transfer occurs when valid = '1' AND ready = '1' on the same clock edge.

CLK   : ___‾___‾___‾___‾___‾___‾
valid :    |‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾|
ready :    |‾‾‾‾‾|_____|‾‾‾‾‾‾‾|
data  :    | D0  |  X  | D1 | D2|
            ↑           ↑    ↑
         transfer  pause  transfers

Avalon-MM vs AXI-Lite Comparison

Criterion	Avalon-MM	AXI-Lite
Standard	Intel/Altera	ARM (AMBA)
Separate channels	No	Yes (5 channels)
Wait mechanism	waitrequest	valid/ready handshake
Native burst	Yes	No (AXI4 full)
Auto-generation	Platform Designer	Vivado IP Integrator
Typical use	Cyclone, Arria, Stratix	Zynq, UltraScale

Key Points

Reset polarity: always check the convention used by the IP and tooling. Many Avalon interfaces use active-high reset (reset = '1'), matching the platform examples.
Read latency: for slaves with RAMs or computations, use readdatavalid rather than waitrequest to indicate data is ready
Byteenable: if the slave doesn't support partial accesses, ignore byteenable and treat all accesses as 32-bit
Platform Designer: in practice, Avalon interfaces are auto-generated by the tool. Understanding the signals enables creating custom IPs correctly

Example: Custom IP in Platform Designer

# Declaring the Avalon-MM interface in the IP TCL file
add_interface s0 avalon slave
set_interface_property s0 addressUnits WORDS
add_interface_port s0 address address 2 Input
add_interface_port s0 read read 1 Input
add_interface_port s0 write write 1 Input
add_interface_port s0 readdata readdata 32 Output
add_interface_port s0 writedata writedata 32 Input
add_interface_port s0 waitrequest waitrequest 1 Output

📝 Test your knowledge - Chapter quiz