Sticks and Stones, page-6

Here are three recent NASA SBIR proposals that have been accepted that may end up involving Akida - I have seen several proposals start in ph1 with a vague description of neuromorphic computing and then specifically mention Akida in ph2. So these are projects I will be watching closely.

https://sbir.nasa.gov/SBIR/abstracts/22/sbir/phase1/SBIR-22-1-H6.22-2264.html

Niobium Microsystems

Scalable Neuromorphic Energy-Efficient Accelerator for Heterogeneous Processor Architectures

Technical Abstract (Limit 2000 characters, approximately 200 words):

Under this effort, Niobium Microsystems, Inc. is proposing a low power computing architecture accelerator for neuromorphic processing which can enable real-time sensor data processing and autonomous decision making that is cost-effective and scalable to the growing data ingestion and processing needs of future autonomous systems. The proposed architecture will be highly scalable and compatible with modern processor systems (such as RISC-V or ARM), so that it can be easily adopted in a variety of new systems, and also easily integrated into existing systems. Additionally, Niobium proposes to integrate the proposed accelerator into a larger SoC that will serve as a proving ground and reference design for the accelerator concept. The SoC will be capable of acting as a primary processor in systems or as a co-processor to existing systems. Ultimately Niobium intended to utilize this accelerator as a standard block in its family of heterogeneous processor architectures.

Niobium proposes the following four technical objectives for Phase I:

(1) Study prior efforts and capture the performance and efficiency metrics as well as the limitations of existing platforms;

(2) Propose a novel architecture for a neuromorphic accelerator compatible with heterogeneous processor platforms (RISC-V- or ARM-based);

(3) Explore available MRAM technology (GlobalFoundries 22FDX), characterize its PPA and propose ways for incorporating into the architecture; and

(4) Estimate performance, power and efficiency metrics for comparison to existing solutions.

Potential NASA Applications (Limit 1500 characters, approximately 150 words):

Space platform which require on-board energy efficient inference capabilities and possibly decision making and action will benefit from the low-power energy efficient inference capability of Neuromorphic processors. Long range missions that will require long-term unsupervised learning and adaptation based on constantly evolving unpredictable conditions can also benefit by the learning modalities that Neuromorphic architectures uniquely support.

Potential Non-NASA Applications (Limit 1500 characters, approximately 150 words):

Niobium is pursuing a fabless semiconductor model & planning to incorporate this accelerator into future energy-efficient SoCs along with existing accelerators for DNNs, cryptography & other computationally intensive functions. These energy-efficient processor SoCs will target energy-constrained application markets (unsupervised sensors & sensor networks, lightweight robotics, drones, wearables).

https://sbir.nasa.gov/SBIR/abstracts/22/sbir/phase1/SBIR-22-1-H6.22-2330.html

Brisk Computing, LLC

Adaptive Neuromorphic Processors for Cognitive Communications

Technical Abstract (Limit 2000 characters, approximately 200 words):

The objective of this work is to develop highly Size, Weight, and Power (SWaP) efficient neuromorphic processors that can train deep learning algorithms. The training phase for deep learning is very compute and data intensive. Being able to train a network on the satellite eliminates the need to send large volumes of data to earth for training a new network. However, this requires an extremely energy efficient deep learning training processor. We will develop resistive crossbar neuromorphic processors, with the primary target being to train deep learning algorithms. Although our system would work for any type of data, we plan to focus on networks for cognitive communication applications. We will also look at processing networks for other data sets. The key outcomes of the work will be the processor design, processor performance metrics on various applications, prototype system, and software for the processor.

Potential NASA Applications (Limit 1500 characters, approximately 150 words):

Potential NASA applications include various deep learning training and inference tasks on satellites. These include cognitive communications, processing sensor outputs, and scientific experiments. Additionally, the developed system could be used for UAVs.

Potential Non-NASA Applications (Limit 1500 characters, approximately 150 words):

The non-NASA market would be primarily for edge processing, where power is highly limited. The market includes both the DoD and the commercial market. DoD applications include cognitive communications, sensor processing, cognitive decision making, and federated learning. Commercial applications include communications systems, automobiles, consumer electronics, and robots.

https://sbir.nasa.gov/SBIR/abstracts/22/sbir/phase1/SBIR-22-1-H6.22-2286.html

Exploration Institute, LLC

Neuromorphic Electronics that Rethinks Verifiable Efficiency on Spacecraft (NERVES)

Technical Abstract (Limit 2000 characters, approximately 200 words):

To meet the NASA need for power efficient algorithms that improve onboard autonomy, Exploration Institute proposes to develop NERVES, an approach for power efficient, verifiable generic calculation and signals processing onboard resource constrained systems using the latest neuromorphic hardware.

Through our substantial experience in applying and developing neuromorphic algorithms for spacecraft systems, we have determined that the performance of our algorithms can be substantially improved if the lowest level substrate, the building blocks, were designed to use the most power efficient traits of neuromorphic hardware. More efficient performance of key mathematical operations in neuromorphic hardware would provide high value to spacecraft developers as they can translate their existing work directly into a vastly more power efficient and faster processing system. NERVES is driven by a practical need and the pipeline to commercialization is already established through Exploration Institute’s track record and current work.

NERVES directly maps conventional algorithms to any neuromorphic processor, combining the benefits of more capable, well known algorithms with the power savings of a neuromorphic architecture. A neuromorphic chip like Intel’s Loihi has a computational power density of more than 1000x that of a CPU or GPU for some tasks. Based on our initial analysis, Exploration Institute predicts that NERVES will enable these kinds of power savings (or conversely, computational capacity increase for the same power) which will greatly improve NASA's capacity for onboard autonomy.

As an added bonus, NERVES provides a more verifiabile approach to neuromorphic computing in space, by allowing the use of verified computing approaches (non-neuromorphic, conventional) on neuromorphic hardware with all the power savings that can entail. This enables more likely adoption and infusion into NASA programs.

Potential NASA Applications (Limit 1500 characters, approximately 150 words):

For a given power budget, using NERVES in concert with neuromorphic hardware, NASA could run significantly more complex processing onboard that will enable more onboard autonomy. With such a general, infrastructure-level additional capability, the potential applications are numerous, including: safer human habitation modules, faster onboard autonomy for navigation and other applications, more automated onboard Fault Management, and a foundation to build onboard cognitive computing to support general operations.

Potential Non-NASA Applications (Limit 1500 characters, approximately 150 words):

NERVES is specifically designed for spacecraft systems such as Gateway, planetary robotics, and government and commercial satellites in general, but also applies to any autonomous system, particularly autonomous vehicles, and is especially useful for applications that are power constrained and mobile (for example: agricultural and automated platforms).