v1.0.0 — First GDSII: FP8 tensor accelerator on 7nm ASAP7

𓁰 PtahCore v1.0.0 — First Silicon

The first release of PtahCore: an open-source FP8 tensor accelerator hardened to real GDSII on ASAP7 7nm, 100% open toolchain.

✅ GDSII Complete (Phase 6)

Block	Area	Freq	Setup Slack	Hold Slack	DRC
mac_cell	750 µm²	250 MHz	+1928 ps	+13 ps	✅ Clean
mac_tile	773 µm²	250 MHz	+1620 ps	+15 ps	✅ Clean
mac_row_abut (32 cells)	71,984 µm²	250 MHz	+180 ps	+48 ps	✅ Clean
mac_grid (32×32)	2.27 mm²	250 MHz	+1156 ps	+435 ps	✅ Clean

🏛️ What's Included

• 21 RTL modules — fp8 decode/encode → mac_cell → mac_tile → mac_row → mac_grid → chip_top
• 89 tests green — 37 Python + 52 RTL cocotb
• Full ISA — 6 instructions (LOAD, MMA, STORE, BAR.INIT, WAIT, REPEAT)
• Dual fp8 — both e4m3 and e5m2 per-instruction
• Async STORE with barrier overlap
• Traveling clock architecture — no array CTS, STA-visible through all 1024 tiles
• OpenROAD flow — full P&R pipeline on ASAP7

🧠 Architecture

Blackwell-shaped: LOAD→MMA(REPEAT)→STORE with distributed TMEM accumulators, output-stationary 32×32 systolic array, and a source-synchronous traveling clock.

📜 License

Apache 2.0.

---

𓁰 The fight with real silicon physics lands here as it happens.