Encharge Recharges In-memory Compute
Startup Encharge AI has exited stealth mode, announcing its in-memory-compute technology. Its chips use analog technology and provide 20x more performance per watt than competitors.
Targeting automotive and surveillance, yet another AI-chip company has entered the market. Encharge AI emerged from stealth mode in December by announcing $21 million in funding and an accelerator chip that promises up to 20x more performance per watt than existing digital technology.
Encharge’s team has come out of Princeton University and published several research papers. The team has also built silicon prototypes over the years that demonstrate the advantages. Its strength is low power consumption through analog compute—which uses older and therefore lower-cost manufacturing technology—and software-development tools that virtualize the hardware for AI. Encharge plans to sample chips this year and start production in 2024.
In-memory compute (IMC) products have already emerged, and several AI-chip start-ups are betting on them as the next-generation acceleration solution. The technology, however, has got traction in niche applications but failed to enter mainstream ones. Its credibility declined when Mythic, which had raised $165 million, cut its staff; that company is now barely hanging on. To survive, Encharge must learn from the successes and failures of others.