What does ‘full custom’ really mean?

Question

It’s common in my intro digital logic/VLSI textbooks to see mention of “full custom” chips versus ASIC chips. I’m interested in understanding the difference between the two in the context where both are used to implement the exact same high-level architecture. The broad strokes distinction seems essentially to be that ASICs are formed by using standard cells whereas full custom designs essentially think about each transistor individually. My question is about at which level of the structured design abstraction (microarchitecture, logic, circuit, and finally physical design) the two approaches diverge. It seems to me that they certainly diverge at the physical design level (we can use specific layouts tricks depending on the situation in adjacent cells etc.). Do they diverge at the circuit level too? That is, might a full custom design of some our fixed architecture use domino logic to eke out advantages over a standard cell using static CMOS? Are there divergences at even higher levels of abstraction?

Answer 1

ASIC = Application-Specific Integrated Circuit.

Standard Cell and Custom are subsets of ASICs.

They diverge at the Synthesis level. The synthesis tool can only use what is available on the silicon Standard Cells. For custom there are almost no limitations.

Note that this technology is constantly improving, I am using the terms that were used for the ASICs that I was involved in.

Standard Cell implementation, AKA sea-of-gates, development is cheaper and faster. The silicon masks are always the same, and they can be pre-fabricated waiting for the custom steps. Only the metal layers will require custom masks.

Custom allows tweaking for maximum performance. But the manpower required can be enormous. I don't know what automation is available now, but in the past a custom processor ASIC could have a hundred or more layout engineers.