The von Neumann era is over. Welcome to neural compute.
Non-Von bridges the gap between AI algorithms and the silicon that runs them. By pairing every processing core with its own memory, our novo architecture eliminates the memory bottleneck and improves performance per watt by an order of magnitude — for AI and multivariate sensor processing at the edge.
AI has outgrown the hardware it runs on.
Processor and memory are physically separate. Every cycle, data must crawl back and forth across the chip — wasting power, generating heat, and throttling AI.
Each core is paired with its own memory, so millions of nodes process in place and in parallel. No commute, no monolithic memory unit, no bottleneck.
For decades, computing has relied on the von Neumann architecture— physically separating the processor from the memory. GPUs were built to push screen pixels and later adapted for AI; they don't fit the way modern models actually compute. AI isn't failing. The hardware architecture running it is.
We didn't build a faster processor. We removed the commute.
Non-Von's architecture is designed for AI from the ground up. We removed the traditional monolithic memory and caches and paired each core with its own memory — inspired by the parallel circuitry of the brain.
Compute next to memory
Data lives where the math happens. Eliminating data shuttling removes the memory bottleneck and radically cuts the power budget.
Sparse-native engine
Traditional GPUs require structured sparsity. The novo architecture processes unstructured sparsity directly — computing only what is necessary.
Datacenter throughput, edge power
The throughput of a data center at the power budget of an edge device — saving money and reducing energy consumption.
Heavy AI, unleashed from the cloud.
By removing the memory bottleneck, Non-Von brings demanding AI and sensor-processing workloads directly to the edge.
Vision & sensor processing
Real-time, parallel processing of imagery and multivariate sensor streams for robotics, autonomous systems, and inspection.
Edge AI & transformers/SLMs
Run sparse, quantized models — from MobileNet to vision transformers — on battery-powered devices without the data center.
Defense & aerospace
Autonomous decision-making in disconnected, power-constrained environments. Work supported by ONR and NAVSEA.
Fielded at the edge and on a drone
Our novo1 proof-of-concept chip was fabricated in 2024 and demonstrated in a fielded commercial product, processing data at the edge.
Numbers that redefine the silicon.
Plug-and-play with the frameworks you already use. Non-Von can take any trained model and implement it as a sparse, compacted system on silicon.
BUILT WITH AND BACKED BY
Stop waiting on the bottleneck. Build what comes next.
Tell us about your workload and we'll show you what the novo architecture can do.