TetraMem
TetraMem develops analog-in-memory computing hardware based on resistive RAM (RRAM), in which calculations are performed where the model weights are stored. The approach targets edge devices, reducing power consumption and enabling real-time AI in products that previously lacked the power budget for it. Its architecture is also positioned to improve thermal efficiency and support high-precision calculations in processors.
- Semiconductor
- Artificial Intelligence
- Hardware
Funding
Funding not disclosed
Founders
Product
Problem
Modern processors face thermal efficiency limits that constrain performance and operating temperature, and high-precision calculations intensify heat management problems in computing devices. Conventional digital architectures also move data between memory and processing units, which consumes power that many edge products cannot afford.
Solution
TetraMem develops analog in-memory computing processors built on resistive RAM (RRAM), an architecture in which computation is carried out directly inside the memory array where the model weights are stored. By processing data in place, the design reduces the energy spent moving data and lowers the power consumption of inference workloads. This allows edge devices to run AI tasks in real time for products that could not previously support them. The company describes its approach as improving thermal efficiency and allowing processors to operate at higher temperatures. Its architecture is also described as enabling high-precision calculations, including arbitrary precision. The company presents this as an inference-era compute platform that brings processing closer to the data.
Target Audience
Primary customers are processor manufacturers and developers of edge computing devices that need energy-efficient, real-time AI inference and better thermal performance.
Features
- Analog-in-memory compute architecture implemented with resistive RAM (RRAM)
- Computation performed where model weights are stored, reducing data movement between memory and processor
- Reduced power consumption for on-device and edge AI inference
- Enhanced thermal efficiency and increased operational temperature range
- High-precision calculation support, described by the company as arbitrary precision