The company's "generally available" Matrix Processing Engine (MPE) is designed to enable high-performance AI processing for IoT and other edge equipment within the smallest energy envelope at an industry-leading ratio. Running AI solutions on an ASIC for domain-specific applications, MPE, says the company, enables the industry's best performance-to-power usage - meaning equipment can operate at high speeds to handle AI quickly, with low energy consumption.
"Balancing the cost-performance-energy equation has been a challenge for developers looking to bring AI-enabled equipment to market at scale," says Dr. Lin Yang, chief scientist, GTI. "The GTI founding team has been watching the industry struggle with this challenge for decades, and believe that our AI Processing in Memory (APiM) and Matrix Processing Engine provide an elegant solution to avoid having to make trade-offs. By deploying APiM and MPE on a standard, commoditized ASIC, GTI is enabling our customers to bring innovative, AI-enabled devices to the masses.”
The company's first-generation product is a 7 x 7-mm 28-nm ASIC that uses just 300 mW to deliver 9.3 TOPS/W that can process audio and video input. With its small size and the ability to combine between two and 32 chips on one board for heavy compute loads or separate task handling, the Lightspeeur 2801S, says the company, enables the broadest range of use cases and equipment designs, at a price point 10X lower than competitive hardware.
"We are paving the way for the next wave of AI products to make it to market," says Kimble Dong, CEO of GTI. "We recognized that device makers were compromising on essential design variables in AI-enabled equipment and have sought to solve this over the past few decades. Our offering marries our "edge-first" approach with ultra-fast AI data processing technology, low power consumption, and a small chip design to enable the best AI experience and performance at a low cost, within any AI use case, physical fit, and deployment."