High performance processors for AI, everywhere
The first truly scalable and reliable analog in-memory computing
We need sustainable computing for AI
Bottlenecked by memory wall (<10 TOPS/W)
No bottleneck (200+ TOPS/W)
Can fit entire model in CIM
Compute-in-Memory (CIM)
Non-volatile memory (NVM)
Breakthrough system design
First product will fit in the palm of your hand and use <1 W
Future products will leverage advanced packaging to scale beyond today's limits
Moore's law is dying. We're unlocking new scaling laws
| Technology | Companies | Power Consumption | Key Limitation |
|---|---|---|---|
| GPU/TPU | NVIDIA, AMD, Google | High power (~1,000 W) | Efficiency blocked by memory wall |
| FPGA | Xilinx, Altera | Medium power (~100 W) | Not optimized for AI |
| Digital CIM | MemryX, D-Matrix | Low power (~10 W) | Low capacity - will hit the memory wall |
| Analog CIM | EnChargeAI, Mythic, Sagence | Low power (~5 W) | A/D conversion bottleneck & inference only |
Providing high-impact solutions from edge to datacenter
Real-time processing in low-network, multi-platform environments
Smart sensors, 5G networks, wearables, infrastructure
Local LLMs, AI Agents, AI PCs, Servers, Data Centers
Our hardware's efficiency is necessary to achieve AGI
