Mythic, an AI processor developer, has adopted two tools from Mentor, a Siemens business, to help it realize its intelligence processing units (IPUs).
The company’s designs use analog computing techniques to perform calculations that fuel deep neural network inference within flash memory arrays.
Ty Garibay, Vice President of Engineering at Mythic explained that the design target sets a number of challenges.
“Mythic IPUs leverage analog computing to perform the calculations required for deep neural network inference inside flash memory arrays. This requires us to simulate thousands of analog-to-digital converters at extremely demanding accuracy specifications,” he said.
“We selected Mentor’s Analog FastSPICE Platform because it can deliver nanometer-scale SPICE accurate results and 5x productivity improvement compared to other solutions. Additionally, the full-spectrum device noise analyses help demonstrate excellent correlation with measured silicon.
“The Symphony Mixed-Signal Platform helped us extend our verification coverage to include the analog-to-digital interface in our IPUs.”