Major programs running with the Navy, the Air Force, DARPA, the Missile Defense Agency, Homeland Security and the Department of Energy, alongside Lockheed Martin and the national laboratories.
Electrochemical recovery and low-flammability composites
An automated screening campaign that searches high-entropy alloy compositions for high-temperature strength and oxidation resistance together, narrowing a very large candidate space down to a shortlist worth making and testing.
Recovers gold, copper and other constituents of value from electronic waste electrochemically and at room temperature. 25% higher recovery, 75% less chemical reagent, 30% lower water and electrical cost, and no toxic emissions. Commercialised as E-RECOV.
Idaho National Lab · Colt Refining
A curated database and learned models relating composite formulation to flammability, so candidate materials can be screened before they are manufactured and burned.
Real-time network defense and source-code vulnerability detection
The programme CyberNeuro-RT came out of. A hybrid supervised and unsupervised deep learning ensemble for network intrusion detection, ported to neuromorphic silicon so inference runs at a fraction of the power a GPU needs.
Pennsylvania State University
Static and learned analysis of source repositories to surface exploitable defects before code ships, trained against mission-critical codebases. Commercialised as CNRT CodeLens.
Lockheed Martin · U. Wisconsin–Madison
Hyperspectral detection of UAVs and concealed contraband
Detects and classifies small uncrewed aircraft from hyperspectral imagery, using spectral signature rather than shape, so targets that defeat conventional optical tracking are still resolved.
Lockheed Martin · Bodkin Design
Identifies concealed contraband by spectral signature at ports of entry, with on-sensor neuromorphic inference so screening runs at the speed of the queue.
Lockheed Martin · Bodkin Design
Metamaterial surfaces for infrared beam control
Engineered metamaterial surfaces on micromirror arrays that steer and shape infrared beams without moving bulk optics, reducing the size, weight and power of the assembly.
Duke University
A sensor that performs part of the recognition task in the optics themselves, using metamaterial elements, so less computation is required downstream of the focal plane.
Lockheed Martin
Multi-sensor object detection and tracking
Fuses several dissimilar sensors into a single track picture, holding identity across handoffs where any individual sensor loses the object.
Ground-based detection on neuromorphic hardware
Detection and tracking of hypersonic threats from ground-based sensors, with inference on neuromorphic hardware to hold the latency budget that the closing speeds demand.
Lockheed Martin · University of Florida
Verification and validation of fielded models
Methods for establishing that a trained model behaves within specification before it is fielded, addressing the assurance gap that keeps machine learning out of safety-critical defence systems.
North Carolina State University
Weapons separation evaluation toolsets
Modelling tools for evaluating store separation from an aircraft, replacing part of a test campaign that is otherwise flown.
Radar and signal processing, real-time synthetic aperture radar simulation
A simulator that generates synthetic aperture radar imagery in real time, so detection and classification models can be trained and evaluated against scenarios that are impractical or too expensive to fly.
Lockheed Martin
We reply from San Jose, usually within two working days.
