Two-Mask Customization

This is one of the most important claims in the whole program. If the upper-mask customization path is viable, the economics of model-specific chips change dramatically.

Key points

This is the bridge between research novelty and business viability.

A weaker version of the idea is still interesting technically, but the two-mask path is what makes the model-specific platform scalable.

The design narrative needs to show why mask customization is more than a manufacturing footnote.

Customization scope

Metal/Via only

Reference material calls out two masks instead of a full mask set.

Design cycle target

~1 week

This is the automation thesis for the front half of the flow.

Tapeout target

~2 months

The repo uses this to frame turnaround expectations.

Why two masks matter

If only the upper metal and via layers need to change between model-specific spins, the program can preserve a stable base die while still encoding new workloads into the physical layout.

What still has to be proven

The repo is already careful here: economic claims are paired with truth-policy work because the manufacturing story is only credible when linked to actual physical-design evidence.

Stable base die assumptions must hold.

RTL and macro strategy must map cleanly into layout.

Fab handoff language must stay synchronized with evidence quality.