ASIC Development
A flow whose last step cannot be undone
Close each question in the stage that can still answer it. Nelix has not taped out, silicon is targeted no earlier than ~2029.
Design targets · not measured silicon results
- Trust boundary
- Supporting logic
- Signed data in flight
Tape-out is where mistakes stop being editable
After masks are ordered, a bug costs a respin. Timing, DFT and physical sign-off each catch failures that more simulation will not.
- Measured epoch
- Flagged for re-measure
- Light = the digest being extended
Gate commitment on closed questions
FPGA validation before RTL freeze; constraint-driven synthesis; STA across PVT; DFT early; physical re-closure on parasitics; sign-off before any mask set. MPW shuttle for first silicon.
Sequence
Order is the mechanism, and nothing here is optional.
- Ordered step
- Only exit
What follows from the diagram
-
FPGA before freeze
Architecture questions closed on a reprogrammable platform
-
STA across corners
Setup and hold closed over PVT, not a nominal point
-
DFT early
Scan and memory BIST so die can be screened
-
Sign-off gate
Timing, PI, DRC/LVS before commitment
Reviewing this mechanism?
The specification can still change. That stops being true after tape-out.