LLM Accelerator — SystemVerilog RTL
HardwareIntegrated a self‑attention head with matrix multiply, GELU, and LayerNorm to build a BERT‑style accelerator; implemented MAC, GELU, accumulate units and a 2D systolic array for high‑throughput inference; added fixed‑point softmax and buffering for full pipeline.
- SystemVerilog
- RTL
- Systolic Array
- FPGA