Acing the Automotive AI Challenge
Automotive OEMs are adding more sensors to their vehicles to support Level 2+ through level 4 ADAS and autonomous driving. Furthermore, the resolution of those sensors continues to increase to improve the accuracy of the detection, recognition, and classification of objects. This in turn drives up the level of AI performance required to implement a given ADAS level. AI accelerators must be able to process hundreds of TOPS in addition to having the flexibility to support a wide range of continuously evolving suites of neural networks. Additionally, because size, weight, and power are critical in automotive systems, AI must also be performed in the smallest possible packages and operate at the lowest possible power levels.
Solutions Built for Automotive
Untether AI created an accelerator architecture based upon an at-memory computation engine to deliver the highest density of compute performance while simultaneously delivering the industry’s leading TOPs / watt. Our AI acceleration family provides up to Peta OPS of processing while also conforming to the small physical and thermal form factor required by vehicles that offer ever-increasing amounts of automation and autonomy. Our products are host processor agnostic and support the neural networks traditionally found in automotive, as well as emerging ones including BEVformer and Vision Transformers.
Our family of acceleration devices is AEC-Q100 compliant and integrate ISO 26262 safety mechanisms to achieve ASIL-B/D functional safety levels. The family supports OTA updates with best-in-class support to make your IP more secure. With up to 250 FP8 TOPs in a 7 watt footprint, the family can easily address AI acceleration for the most demanding L2+ / L3 ADAS applications and with a thermal and power footprint that is otherwise unheard of at these performance levels.See our AI inference devices