RTL-to-GDS implementations of blocks from my research design, a scaled-down NVIDIA-style GPU first built on a Virtex-7 FPGA. Each block goes through logic synthesis (Synopsys Design Compiler), place-and-route (Cadence Innovus) and sign-off timing in the FreePDK45 / OSU 45 nm standard-cell library.
| Block | Status | Highlights |
|---|---|---|
tensor/ — tensor-core sub-partition (16 int16 MACs) |
Signed off | 400 MHz post-route (setup +45 ps, hold +229 ps, OCV), 0 DRC, 0.14 mm² die; synthesis sweep to 1.25 GHz |
sm/ — full SM (4 tensor cores, operand buffers, 2 clock domains) |
Planned | Multi-clock SDC and CDC constraints |
Tools: Synopsys Design Compiler · Cadence Innovus · Tcl · Verilog
Library: FreePDK45, OSU gscl45nm standard cells
