Sagence AI develops analog in-memory compute technology that delivers high-performance AI inference with 100X lower power consumption and 20X lower costs compared to traditional digital solutions. This approach addresses the limitations of increasing digital chip densities and energy demands, making AI more economically viable and sustainable for widespread applications.
Funding
Funding not disclosed
Founders
Product
Problem
The increasing computational demands of AI inference, particularly for generative AI models, are driving up energy consumption and costs, hindering widespread adoption and raising sustainability concerns. Traditional digital computing architectures struggle to efficiently handle the massive matrix multiplications required for AI inference, leading to performance bottlenecks and high power usage.
Solution
Sagence AI offers an analog in-memory compute solution that significantly reduces the power consumption and cost of AI inference. By performing computations directly within memory cells, Sagence AI eliminates the need to move data between memory and processing units, resulting in substantial energy savings and improved performance. This approach enables the deployment of AI models in a wider range of applications, from data centers to edge devices, while minimizing environmental impact and maximizing economic viability. The technology achieves this through deep subthreshold compute inside multi-level memory cells.
Target Audience
Sagence AI targets businesses and organizations deploying generative AI models, including those in content creation, personalized experiences, anomaly detection, and security, who require high-performance, energy-efficient, and cost-effective inference solutions.
Features
- Analog in-memory compute architecture for efficient AI inference
- 100X lower power consumption compared to traditional digital solutions for MAC functions
- 20X lower costs due to the use of mature, reliable low-cost process nodes
- Integrated storage and compute on a single device, reducing complexity and power consumption
- Deep subthreshold compute inside multi-level memory cells
- Deterministic, programmable architecture for real-time defect detection
- Scalable from data center to edge applications