American chip startup Cima has released a multimodal edge artificial intelligence (AI) chip. Recently, demand for AI chips has expanded from existing data center GPUs to chips for on-device AI or edge AI, and competition amongst edge semiconductor corporations to develop them can be intensifying.
VentureBeat reported on the tenth (local time) that Cima had released its second-generation multimodal edge AI chip, ‘MLSoC Modalix’. This chip used TSMC’s 6nm process.
MLSoC Modalix is a single edge AI platform that supports all frameworks, networks, models, sensors, and all modalities for edge AI applications, including audio, voice, text, and image.
It’s scalable from 25 to 200 TOPS depending on the applying, and combines an integrated image signal processor (ISP), PCIe Gen 5 support, and eight Arm Cortex-A65 CPUs to handle various AI tasks. It also provides a code-less software tool called ‘Palette’ that makes it easy to write down software for MLSoC chips.
“This product is ideal for industrial automation, robotics, healthcare, smart vision systems, aerospace and defense, and anywhere a multimodal element is required,” said Krishna Rangasai Seema, CEO.
Specifically, MLSoC Modalix delivers 10x higher frames per second per watt or inference performance per second per watt than NVIDIA’s edge AI chips, Orin or Xavier.
“We’ve got all the time been 10x higher than our immediate competitors in real-world applications, and this time we expect that gap to widen to well over 10x,” said CEO Rangasai.
Reporter Park Chan cpark@aitimes.com
