Logic Die

Awareness, Perception & Cognition

Near Memory Fabric

Move only what matters

Memory Die

Refresh-free transient memory

FotoNation’s perceptive silicon designed to deliver human-like cognition with an ultra- efficient, low-power design.

Features

Low Power

Built to bring humanoid-class perception and intelligence under 50W, versus around 150W for today's centralized subsystems.

Always On

Continuous sensing and inference without waking the full system. Ultra-low-power detection of audio, video, and motion events keeps devices aware around the clock, so battery life isn't the cost of vigilance.

Intelligent

The Programmable Core Neural Network (PCNN) delivers full DNN, CNN, and transformer support — not fixed-function inference. Deep scene understanding, multimodal fusion, and reasoning turn raw sensor data into decisions, not just detections.

Modular

Built on a stackable architecture — compute and memory integrate through a near-memory fabric and hybrid-bonded memory, rather than communicating over a conventional bus. Each layer connects directly, eliminating the data-movement bottleneck that limits traditional chip designs.

Scalable

The same architecture scales from single-die configurations to full 3D chiplet arrays, growing compute and memory bandwidth together without a redesign. One foundation stretches from smart cameras to high-throughput Physical AI compute.

Adaptive

Elastic ISP and DeVIS balance image quality, power, and responsiveness in real time — tuning processing to the scene and the power budget rather than running one fixed pipeline regardless of conditions.

Multi-modal

Fuses vision, audio, thermal, and motion into a single perceptual pipeline. Sensor-agnostic architecture means one platform handles camera, mic, and IMU data together — not as separate bolted-on subsystems.

Human-like

Perception modeled on how attention actually works: processing only what matters, ignoring the rest. Context-aware, intent-driven understanding — closer to biological efficiency than brute-force compute.
In Action

The Compute Foundation for Next-Gen Robotics

Current robots waste 150W+ on perception and compute, severely limiting battery life and reliability.Our architecture is built for Physical AI, bringing that same class of perception under 50W by processing only essential sensor data.

Powered by our near-memory fabric and TRAM™ to eliminate data-movement bottlenecks, the result is real-time, multi-modal fusion without the thermal throttling of conventional chips. This is the compute foundation robotics OEMs need to move from impressive demos to deployable economics.

Get in touch to know more
In Action

Real-Time Awareness for Drones and Vehicles

Drones and vehicles need continuous awareness within tight weight, power and battery budgets. TriSilica® fuses camera, audio, motion and other sensor data on the device and passes compact latent information to flight and driving systems instead of streaming raw data. In vehicles, the same always-on perception supports sentry mode, driver monitoring, in-cabin AI and ADAS perception.

Get in touch to know more
In Action

Personal Agentic AI in an Edge Box

Agentic AI in the home needs to be always aware and instantly responsive. As AI in an Edge Box, TriSilica® runs perception and context locally for personal assistants, home intelligence and household management, engaging larger AI compute only when it adds value.

Get in touch to know more
Performance

TriSilica

Other

Designed for 10–100× lower power than incumbent edge AI platforms*, with targets of more than 60 TOPS/W and around 0.3 pJ/bit.

TriSilica

Other

[Copy required about 4-50x memory bandwidth]
*All figures are target specifications based on internalarchitecture modelling and prior siliconimplementations; results vary by workload, precision,process node and system configuration.

Additional Data & Research

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.
If there is any particular information that you are looking for but are unable to find on this page, get in touch with our team and we’ll try our best to help you.

Get In Touch

Minimal data is collected, stored securely, and never shared without consent unless required by law.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.