Last month we saw Infinisim announce a new SPICE simulator and this month it’s a company called Gemini.
What’s new with Gemini is that they exploit multi-thread and multi-core to a greater extent. For understandable reasons the old-guard SPICE tools (HSPICE, Spectre, Eldo) have been slow to re-architect to fully exploit multi-thread and multi-core. It’s plain easier to re-write a tool from scratch rather than try and cobble in fundamental change.
Gemini also had the marketing insight to use 3D graphical charts to help position their new efficiency:
They also had the good sense to quote their customers instead of themselves.
Having Jim Solomon involved with the company is a huge benefit, and I expect to see many more customer success stories coming from this start-up in the next year.
The only issue that I have with all of these SPICE claims is the one about absolute accuracy. They seem to be saying that you can perceive a difference between 99% and 99.9% accuracy with respect to SPICE. If process variation within an IC is getting larger with each new node then how would a SPICE simulator show that to me?
Arguing about how accurate your SPICE simulator is kind of misses the mark because silicon variation is much larger than any modeling accuracy. SPICE isn’t reality, silicon is reality. Tell me how close your SPICE simulator is to silicon, thank you.