FlashNPU
Inference accelerator RTL
Runs neural network layers straight out of flash with no DRAM. 32-lane INT4 x INT8 GEMV engine fed by 4 striped flash channels. Bit-exact against a Python golden model.
Recent work outside my internships
Inference accelerator RTL
Runs neural network layers straight out of flash with no DRAM. 32-lane INT4 x INT8 GEMV engine fed by 4 striped flash channels. Bit-exact against a Python golden model.

SKY130 · TinyTapeout analog
Differential 8-bit, 1 MS/s SAR with a monotonic-switching cap DAC. Golden model hits 7.9 ENOB. Bootstrapped sampling switch reaches 78 dB SFDR.
SKY130 · TinyTapeout digital
Baseband and RF switch controller for a BLE backscatter tag. Full P&R at 64 MHz with +3.08 ns slack and 0 DRC/LVS errors.
Interactive tool
Link budget, Smith chart load states, image rejection, sideband spectrum, and bistatic coverage for an N-phase backscatter tag. Runs in the browser.
Hardware I've built and tested
Roles I'm targeting for my engineering career
Most recent first
Selected for competitive industry and university programs