We built the world's first single-transistor neuron and synapse. Drop-in compatible with any CMOS foundry. No exotic materials. No new fabs. Just standard silicon, engineered to think.
Every chip running every AI system today was designed for one job — switch on, switch off. The mismatch between that design and the work we now ask it to do is the central engineering problem of this decade.
One device. Memory and compute in the same place. Standard CMOS, no new materials. The three problems collapse into a single primitive.
Conventional silicon treats switching as binary. We discovered a regime where the same MOSFET, the building block of every chip ever shipped, produces spiking dynamics identical to a biological neuron.
The physics: sub-threshold integration, leaky firing, refractory recovery. All of it, from one device. No capacitors, no resistive memory, no memristive stack.
The implication: neuromorphic computing is no longer a materials problem. It's an architecture problem. And it's solvable, today, at any foundry.
Every aspect of the technology is designed to collapse what neuromorphic hardware typically takes dozens of components to achieve and into the footprint of a standard MOSFET.
Sub-threshold integration plus leak plus refractory recovery: the full LIF (leaky integrate-and-fire) dynamic, produced by the MOSFET's own physics. No external capacitor. No RC network. The neuron is the transistor.
No exotic oxides. No memristive materials. No 3-D stacking requirements. The device is fabricated in a standard CMOS process. 130 nm, FinFET, nano-ribbon, CFET. If a foundry can run logic, it can run us.
Full SPICE model across the operational tuning space. Chip designers simulate and place Newmorphic cells using the same flow they already use for every other block: Cadence, Synopsys, industry-standard EDA. No new workflow.
The endurance plot every other neuromorphic platform fails. Same axes, same scale, side by side. Across nine decades of cycles.
The same device, in many shapes. The substrate is general-purpose; this is where it begins.
Tonotopic wake-word, speaker ID, and in-ear noise cancellation running below the noise floor of today's DSPs.
Event-based sensing for drones, robotics and vehicles. Latency-first inference with biological efficiency.
Continuous biosignal analysis: ECG, EEG, EMG. On devices that run for weeks on a single coin cell.
Decision-ready telemetry from distributed sensor meshes. No backhaul, no cloud dependency.
Standard CMOS heritage means mature rad-hard process compatibility for on-orbit and tactical edge.
Real-time control for robotics, vehicles and drones. Decisions made on the device, at the speed the device needs them.
We believe the next decade of AI hardware will be decided not by who invents the best new material, but by who can deliver the most efficient compute through the fab lines that already exist.
Technical deep-dives, silicon measurements, and field notes from building the first foundry-native neuromorphic primitives.
Whether you're a potential investor, foundry partner, research collaborator or journalist, tell us who you are and what you're working on. We read everything.
info@newmorphic.com