Sagence AI

About Sagence AI

Analog Inference, which now operates as Sagence AI, is a semiconductor company established in 2018 to address the power consumption and cost challenges of artificial intelligence (AI) inference. The company was founded by Vishal Sarin, an expert in analog and memory technologies with over 100 patents to his name. Sarin's background includes leadership roles in analog and AI technology development at major companies like Micron, which directly informs the company's core mission. The executive team is further strengthened by CEO Carey Kloss, who joined in 2020 and brings experience from his role as employee #1 at Nervana Systems and leading AI product engineering at Intel.

The firm is focused on the AI acceleration market, specifically for edge computing and data centers. Its business model centers on designing and commercializing AI inference accelerator chipsets that offer a significant performance-per-watt advantage over traditional digital computing solutions. These chips are designed for a wide range of applications, including smart vision, speech interpretation, robotics, medical devices, and large language models (LLMs). The company targets clients such as mobile device OEMs, surveillance companies, and entities that require extensive video analytics at the edge.

Analog Inference's core product is an AI co-processor chip that utilizes analog in-memory computing. This technology merges data storage and computation within memory cells, which circumvents the need for power-intensive SRAM and DRAM by storing neural network weights in non-volatile, multi-level memory. This approach drastically reduces power consumption, with the company claiming a potential 100-fold decrease for the multiply-accumulate (MAC) function, a fundamental operation in AI. By using mature, low-cost manufacturing processes, the technology aims to deliver performance comparable to high-end GPUs but at a fraction of the cost and physical footprint, which is a key selling point. The company has stated its solution can provide up to 10 times lower power and 20 times lower price and rack space compared to conventional GPU solutions for tasks like running the Llama2-70B model.