Neuromorphic chip for sensing, situational awareness, and decision making in radiative environments (NCS-SBIR SEQ)

Neuromorphic processor with extreme energy efficiency and radiation tolerance. The technology provides upto 40 AIML TOPS (trillion compute operations per second) per watt, for complex Artificial Intelligence needed for autonomy in mission-critical applications in space environments. The processor is desgined to operate as a co-processor for radiation tolerant CPUs, providing more than 100x increase in AIML ops for inference only. The co-procesor communication is through a standard high-bandwidth bus, expected to be PCIe, providing both data and control. In operation, the co-processor logically appears as part of external memory to a CPU, with locations for input and output. The processor consists of two parts, one being an analog component (AVM) for densely connected neural layers, and one being a digital component (DVM) for convolutional or sparesely connected neural layers. The analog component is being built with path-breaking memristor devices that have extreme radiation tolerance. The digital component is being built on 22nm FDSOI (silicon on insulator) that has natural radiation tolerance and environmental robustness. DVM is designed to be exceptionally efficient for processing streams of correlated inputs, such as video feeds. The AVM and DVM parts are tied together with an FPGA that also provides the external interface. A software tool chain enables neural net models to be readily compiled to the hardware neuromorphic processor. The software tool chain takes as input one or more neural nets in a standard ONNX format, with the different layers assigned to DVM or AVM, with the FPGA providing communication between the two.The radiation tolerance will meet or exceed that of space qualified rad-tolerant CPUs.

​Artificial Intelligence processor for perception and autonomous decision making that is suitable for any space mission where the computing subsystem needs to be power efficient. The throughput and power efficiency of the processor makes it usable for low power smart sensors, for sensor and image noise reduction, for super-resolution, and for sensor fusion. The ability to efficiently process video streams from pixels to classified moving objects and reconstructed 3D surfaces makes the processor suitable for mission phases ranging from proximity operations to terrain relative navigation to autonoous roving. The processor is suitable to run neural based anomaly detection, fault classification, and recovery for subsystems ranging from vehicle health to life support to communication. The technology is scalable, and more computing elements can be incorporated on the chips as size and throughput requirements increase. The current technology development is aimed towards inferencing for convolutional neural nets (CNNs), as the technology progresses, it could become a hardware platform for in-situ generative AI for space missions.

This is early/mid-stage (TRL 6) — the most realistic path in is NASA SBIR/STTR, which funds small businesses and research institutions to develop technology aligned with NASA's needs (equity-free, phased funding). Check whether a current SBIR/STTR solicitation topic overlaps with this project's technology area, or contact the project directly (above) to ask.

None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.