Anyone heard of Silicon Telemtry?
I'm highly skeptical of the marketing noise around this. Are people actually putting these IPs into production, or are they just glorified ring oscillators?
A few questions for anyone familiar with silicon telemtry:
-Is there a specific class of monitor IP that handles this without a massive area overhead? Can it actually differentiate between global process variation and localized aging in-situ?
-If we are pulling telemetry from thousands of points across a die, where is this data actually being stored? processed?
-Is this something that's going to become a requirement for high-rel automotive/datacenter, or are we just being sold early-adoption vendor hype?
I'm 20 years in DFT and silicon bring-up, and my philosophy has always been pretty straightforward: run rigorous ATPG, characterize the hell out of the first silicon, set a sane Vmin guardband based on HTOL, ship the product, and sleep well. Got a task from management to look into other ways to deal with reliability without increasing guard bands.
Thank you.