Something which had slipped under my radar (if you'll pardon the expression) - the USAFRL/RTX micro-Doppler SBIR project is intended to use Akida 2/TENNs, following RTX's in-house tests with Akida 1 CotS. As the AKIDA 2 ASIC does not exist at the moment, the project may have used the FPGA version. If memory serves, the FPGA is 8 times slower than the ASIC. Despite this, there is the intention to produce the commercial version for see-in-the-dark radar.
Wait til they get the Akida 2 SoC ASIC.
https://www.defenceconnect.com.au/i...n-contract-with-air-force-research-laboratory
11 December 2024
BrainChip will partner with subcontractor(s) to develop comprehensive set of algorithms and neural networks optimised for BrainChip neuromorphic hardware.
The contract, granted under the US federal government’s Small Business Innovation Research (SBIR) program over the 12-month term of the agreement, is expected to develop neuromorphic radar signalling processing.
The SBIR contract award, under the topic number AF242-D015, is titled “Mapping Complex Sensor Signal Processing Algorithms onto Neuromorphic Chips”.
“Radar signalling processing will be implemented on multiple mobile platforms, so minimising system SWaP-C is critical,” said Sean Hehir, CEO of BrainChip.
“This partnership to improve radar signalling applications for AFRL showcases how neuromorphic computing can achieve significant benefits of low-power, high-performance compute in the most mission-critical use cases. This award is a very strong endorsement from leading organisations such as Air Force Research Laboratory for our industry-leading TENNs offering.”
The contract is an expansion of efforts after a multinational aerospace and defense customer successfully demonstrated radar processing algorithms capable of running on BrainChip’s commercial off-the-shelf neuromorphic hardware as part of an internal research and development initiative.
This current program, however, will develop algorithms based on BrainChip’s proprietary state space model algorithm framework known as TENNs (Temporal Event Neural Network) and will be optimised to run on Akida 2.0 hardware.
The BrainChip TENNs algorithm, combined with Akida 2.0 technology, has successfully demonstrated the capability to run models very efficiently, resulting in significantly higher performance at ultra-low power relative to traditional accelerators running traditional models.
BrainChip’s neuromorphic technology improves the cognitive communication capabilities on size, weight, power and cost-constrained platforms such as military, spacecraft and robotics for commercial and government markets.
The project focuses on a specific type of radar processing known as micro-Doppler signature analysis, which offers unprecedented activity discrimination capabilities. BrainChip is currently in negotiations to enter into a subcontractor agreement with the previously mentioned aerospace and defence company for the completion of the contract award.
BrainChip will partner with the subcontractor to provide research and development services developing and optimising algorithms for a fixed fee totalling $800,000 over the same period.
No other material conditions exist that must be satisfied for the agreement to become legally binding and to proceed. Air Force Research Laboratory will begin making milestone payments in January 2025.
Periodic payments will continue throughout the year concluding in February of 2026.
At the 9 minute mark in the roadmap Jonathan Tapson mentions that AKIDA 2 is available as an FPGA;
https://brainchip.com/brainchip-technology-roadmap/
@ 23 minutes - Akida 3 FPGA Q1 2026!
Akida 3 is more versatile than Akida 1 or 2 which use state machine architecture, whereas GenAI & Akida 3 use instruction set architecture which makes the architecture much more flexible.
Now that the Roadmap is about 8 months old, I found it encouraging to review the video. A few things that were a year out are now just over the next hill. It is packed with groundbreaking advances.
Wait til they get the Akida 2 SoC ASIC.
https://www.defenceconnect.com.au/i...n-contract-with-air-force-research-laboratory
BrainChip engaged under US$1.8m contract with Air Force Research Laboratory
Industry11 December 2024
BrainChip will partner with subcontractor(s) to develop comprehensive set of algorithms and neural networks optimised for BrainChip neuromorphic hardware.
The contract, granted under the US federal government’s Small Business Innovation Research (SBIR) program over the 12-month term of the agreement, is expected to develop neuromorphic radar signalling processing.
The SBIR contract award, under the topic number AF242-D015, is titled “Mapping Complex Sensor Signal Processing Algorithms onto Neuromorphic Chips”.
“Radar signalling processing will be implemented on multiple mobile platforms, so minimising system SWaP-C is critical,” said Sean Hehir, CEO of BrainChip.
“This partnership to improve radar signalling applications for AFRL showcases how neuromorphic computing can achieve significant benefits of low-power, high-performance compute in the most mission-critical use cases. This award is a very strong endorsement from leading organisations such as Air Force Research Laboratory for our industry-leading TENNs offering.”
The contract is an expansion of efforts after a multinational aerospace and defense customer successfully demonstrated radar processing algorithms capable of running on BrainChip’s commercial off-the-shelf neuromorphic hardware as part of an internal research and development initiative.
This current program, however, will develop algorithms based on BrainChip’s proprietary state space model algorithm framework known as TENNs (Temporal Event Neural Network) and will be optimised to run on Akida 2.0 hardware.
The BrainChip TENNs algorithm, combined with Akida 2.0 technology, has successfully demonstrated the capability to run models very efficiently, resulting in significantly higher performance at ultra-low power relative to traditional accelerators running traditional models.
BrainChip’s neuromorphic technology improves the cognitive communication capabilities on size, weight, power and cost-constrained platforms such as military, spacecraft and robotics for commercial and government markets.
The project focuses on a specific type of radar processing known as micro-Doppler signature analysis, which offers unprecedented activity discrimination capabilities. BrainChip is currently in negotiations to enter into a subcontractor agreement with the previously mentioned aerospace and defence company for the completion of the contract award.
BrainChip will partner with the subcontractor to provide research and development services developing and optimising algorithms for a fixed fee totalling $800,000 over the same period.
No other material conditions exist that must be satisfied for the agreement to become legally binding and to proceed. Air Force Research Laboratory will begin making milestone payments in January 2025.
Periodic payments will continue throughout the year concluding in February of 2026.
At the 9 minute mark in the roadmap Jonathan Tapson mentions that AKIDA 2 is available as an FPGA;
https://brainchip.com/brainchip-technology-roadmap/
@ 23 minutes - Akida 3 FPGA Q1 2026!
Akida 3 is more versatile than Akida 1 or 2 which use state machine architecture, whereas GenAI & Akida 3 use instruction set architecture which makes the architecture much more flexible.
Now that the Roadmap is about 8 months old, I found it encouraging to review the video. A few things that were a year out are now just over the next hill. It is packed with groundbreaking advances.
