soNN

=======================

A Reconfigurable Accelerator for Deep Convolutional Neural Networks implemented by chisel3.

How to get started

Dependencies

JDK 8 or newer

We recommend LTS releases Java 8 and Java 11. You can install the JDK as recommended by your operating system, or use the prebuilt binaries from AdoptOpenJDK.

SBT or mill

SBT is the most common built tool in the Scala community. You can download it here.
mill is another Scala/Java build tool without obscure DSL like SBT. You can download it here

Usage

Run this, to check your environmrnt:

sbt 'testOnly sonn.PEArrayConv2Tester'

You will get output like this:

[info] [14.390] ===============ERROR: 0======================
Enabling waves..
Exit Code: 0
[info] [14.394] RAN 7096 CYCLES PASSED
[info] - should create a vcd file from your test
[info] Run completed in 1 minute, 20 seconds.
[info] Total number of tests run: 1
[info] Suites: completed 1, aborted 0
[info] Tests: succeeded 1, failed 0, canceled 0, ignored 0, pending 0
[info] All tests passed.
[success] Total time: 81 s (01:21), completed Jun 19, 2021 3:43:08 PM

Now you can find test_run_dir/PEArrayConv2 folder, use gtkwave or verdi to open PEArrayConv2.vcd in this folder.

Run this, to get Verilog code:

sbt 'test:runMain sonn.getVerilogPEArray'

You can find Verilog code in folder test_run_dir/PEArray.

ToDo

• 设计单个的 PE 单元，实现一维卷积运算；

• PE 阵列设计；

• 仿真测试；

• 总线接口设计；

• 软件驱动设计；

• 上板测试；

File Tree

.
├── build.sbt               # 编译脚本
├── build.sc
├── LICENSE
├── paper                   # 设计报告
├── pic
├── README.md
├── behavioral-level-matlab # matlab 行为级仿真代码
├── src                     # chisel 代码 
│   ├── main
│   │   ├── resources       # verilog 代码
│   │   └── scala
│   │       ├── adder
│   │       ├── myutil
│   │       └── sonn        # 主要 chisel 代码
│   └── test
│       └── scala
│           ├── adder
│           ├── myutil
│           ├── simulator
│           └── sonn        # 测试代码
└── test_run_dir
    └── PEArray						  # chisel 生成的 verilog 代码

Details

设计报告