Mythic utilizes analog FastSpice and Symphony mixed-signal platforms
Improvement of ADC accuracy required for DNN inference within flash memory arrays

Mythic, an AI processor manufacturer, adopted Mentor Siemens Business's Analog FastSpice platform for custom circuit verification and device noise analysis.

Mythic used Mentor's Symphony Mixed Signal Platform to verify the capabilities of the analog-digital logic integrated into the IPU.
Mythic IPUs utilize analog computing to perform the calculations required for deep neural network (DNN) inference within flash memory arrays, and to do this, they must be able to simulate thousands of ADCs with high accuracy.

Mentor's analog FastSpice platform provides nanometer-level SPICE accuracy while offering five times higher productivity compared to other solutions. It can verify analog, radio frequency (RF), mixed-signal, memory, and custom digital circuits. It is foundry certified up to 5nm and provides nanometer-level accuracy that is twice as fast as parallel SPICE simulators.

In addition, it includes comprehensive full-spectrum device noise capabilities to ensure silicon accuracy, and its ability to analyze device noise across the entire spectrum demonstrates excellent correlation with the measured silicon.

The Symphony Mixed Signal Platform is an industry-standard HDL simulator built on Analog FastSpice solutions. It provides fast and accurate mixed-signal verification and validates complex nanometer-level mixed-signal ICs through an intuitive user model, powerful debugging capabilities, and configuration support.

“The use of our platform for computation within Mythic’s analog memory tests the limits of AI hardware architecture verification,” said Ravi Subramanian Mentor, General Manager of Siemens Business IC Verification Solutions. “We expect to achieve aggressive time-to-market goals through this new verification technique.”