Congrats to Mythic on their $70M Series C round, co-led by BlackRock and Hewlett Packard.

Their first chip, the M1076 Mythic AMP (Analog Matrix Processor), implements neural networks in an analog flash memory array, achieving a peak of 35 trillion operations per second using just 3.5 watts in a single 40nm chip.

For a sense of how radically different this is from any digital chip, Mythic can implement an 8-bit multiply and add in a single transistor. Analog compute, in memory, like our cortex.

“Constellation Research Inc. analyst Holger Mueller told SiliconANGLE that it’s well-known that analog chip designs have a big advantage in terms of energy consumption, heating and the physical space they take up. “It’s not surprising that Mythic has scored this round of funding, as there is ample room for these kinds of AI inference chips in ‘internet of things’ and edge devices,” he said.

“We have had strong interest in high-performance video analytics applications such as video surveillance cameras for object detection and classification,” he said. “Other applications include industrial machine vision for high-resolution inspection, defect detection, augmented reality applications for low-latency pose and object detection, and drone applications like object detection, depth estimation and classification.”

VentureBeat news today, and SiliconAngle

2 responses to “Mythic Series C — Analog AI to feed Roko’s Basilisk”

  1. Implementing the future… as laid out in IBM’s 2019 Forecast: Keep in mind that a 65x transistor density advantage can still be had when Mythic moves from 40nm to a modern 5nm fab.

  2. It’s been catching on. From today’s news on the energy efficiency of Analog AI: "Startups including Mythic focus on analog AI using electronics – which Wright says is a ‘great step, and it is probably the lowest risk way to get into analog neural networks.’”

Leave a Reply

Your email address will not be published. Required fields are marked *