Tiny TPU is a lightweight, open-source Tensor Processing Unit (TPU) built for learning, experimentation, and research.
It demonstrates how AI accelerators perform tensor operations such as matrix multiplication using parallel hardware units — inspired by real-world TPUs but simplified and transparent.
This project is ideal for:
- Students learning computer architecture
- AI + hardware enthusiasts
- Open-source contributors
- Resume-worthy system-level projects
Big tech companies keep accelerator designs closed.
Tiny TPU opens the black box.
✔ Understand how AI chips work
✔ Combine ML + hardware knowledge
✔ Learn systolic & MAC-based designs
✔ Open-source and beginner-friendly
- ⚙️ Matrix Multiply Engine (MAC-based)
- 🧮 Fixed-point arithmetic (8/16-bit)
- 🔄 Parallel processing units
- 🧠 AI-oriented dataflow
- 📦 Modular RTL design
- 🧪 Fully simulation-ready
+-------------------+
| Input Buffer |
+-------------------+
|
v
+-------------------+
| Matrix Multiply |
| Engine (MACs) |
+-------------------+
|
v
+-------------------+
| Accumulator & |
| Output Buffer |
+-------------------+
| Layer | Technology |
|---|---|
| RTL Design | Verilog / SystemVerilog |
| Simulation | Icarus Verilog |
| Waveform | GTKWave |
| Control | FSM-based |
| Verification | Python (optional) |
tiny-tpu/
├── rtl/
│ ├── mac_unit.v
│ ├── matrix_engine.v
│ ├── controller.v
│ └── top.v
│
├── testbench/
│ └── tb_top.v
│
├── scripts/
│ └── simulate.sh
│
├── docs/
│ └── architecture.md
│
├── LICENSE
└── README.md
- Icarus Verilog
- GTKWave
- Linux / macOS
git clone https://github.com/Sam-bot-dev/tiny-tpu.git
cd tiny-tpu
Lead Dev
Bhavesh
🌐 GitHub
|