A forest looking straight ahead at a path through some redwood trees

A monolithic platform built to push the physical limits of intelligence

Our unique approach combines the advantages of three key technologies: semiconductors, superconductors, and photonics

A sketch drawing of a superconductor
A sketch drawing of a superconductor

Superconductors for computation

Superconducting electronic devices such as Josephson junctions have unparalleled speed and energy efficiency. These devices exhibit ideal properties for the computations that support AI.

Photonics for communication

Neurons in the brain make thousands of connections. Reaching that level of connectivity with electronics is infeasible. Photonic communication makes it possible. Using multi-planar photonic waveguides on wafers and fiber coupling between wafers, we avoid the communication bottlenecks experienced by conventional neural networks.

A sketch drawing of photonic commuication waves
A sketch drawing of photonic commuication waves
A sketch drawing of a semiconductor
A sketch drawing of a semiconductor

Semiconductors for integration

Silicon bridges the gap between electrons and photons and is a versatile system for many functions including control logic, amplification, light emission, and more. Transistors and light-emitting diodes play a vital role in our approach.

Looking Forward

The path from demonstration to transformation

A cortex on a chip

The essential architecture of a brain, distilled into a compact network at the chip scale. We'll focus on experimentally demonstrating the unique algorithms required to train and inference on sophisticated analog neural networks

Wafer-scale networks

At the next stage of our technology, the network will span an entire 300 mm wafer, leading to new and emergent capabilities. A single wafer performs more synapse operations per second than the human brain

Multi-modular cognitive systems

By tiling wafer-scale networks with an advanced interconnection system, we'll enable the construction of key brain modules (thalamus, hippocampus, neocortex, and more). This level of integration fully captures the essence of our system architecture

Hypercognition

With further scaling, we can construct neural supercomputers more complex than the human brain, operating up to a million times faster. Problems that seem intractable now will be within reach for the first time in history