SBIR-STTR Award

Artificial Intelligence (AI) is driving the fourth industrial revolution as well as permeating every aspect of our day-to-day life. From big data analysis to language analysis and real time translation, from speech recognition to image recognition. The latter is a powerful and quite general application with a scope that spans from medical imaging to autonomous driving and to military applications.Mentium Technologies Inc., spun from a UC Santa Barbara research lab in the Electrical and Computer Engineering department is committed to embrace the AI revolution strong of the experience of its team in the neuromorphic hardware for AI. Indeed, we will develop a neuromorphic chip able to do higher than real-time image recognition and/or object classification on board the UAS. The chip will use 1/100th of the energy while reaching 100x in speed compared to state of the art. The team already had demonstrated 1000x and 1/1000th energy consumption in a smaller scale experimental demo. From this experience UCSB has a patented technology licensed by Mentium Technologies Inc. thanks to this technology and its develpment within this project, the Neuromorphic Chip will empower the UAS with Cognitive functions enabling autonomous guidance, decision making and complex image processing, while keeping the power consumption low.

Potential NASA Commercial Applications:
(Limit 1500 characters, approximately 150 words) Compared to conventional and classical signal processing algorithms, the process done in Deep Neural Networks (DNNs) resembles more to what is happening in human brain and because of that, these networks can provide more useful insight and perception form the surrounding environment. Moreover, by following a teacher like a crew member, they can learn by themselves to how to react autonomously in different and complex situations. All these properties make them a valuable technology to help NASA automates earth and space missions. Here are some of the applications of DNN in NASA-related missions: -Aircraft control-Damage-adaptive decision making-Detect extreme weather in climate datasets-Classification of aerial images-Fire detection and control-Autonomous driving of vehicles for space missions- Automatic feature extraction from large datasets of probes images

Potential NON-NASA Commercial Applications:
(Limit 1500 characters, approximately 150 words) Deep neural networks are envisioned to revolutionize the field of machine learning and their applications because they provide a simple platform to achieve performances beyond what could have been achieved with conventional digital Von Neumann architectures. In fact, despite of being a very young technology, it is already in use in so many commercial applications including but not limited to:-Google is using DNNs for speech recognition in Alexa, for image recognition to diagnose diseases from medical images, for object detection in its self-driving cars, for translating text from one language to another one, etc.-Baidu is using DNN to automatically convert speech to text in mobile phones-The hand-written text on all checks and envelops are automatically read with DNNs-In the huge market of advertising, DNN are now helping to identify the potential costumers for a particular product-All face recognitions and people identifications happening in Facebook pages are possible only because of DNNs-Automatic game playing-Automatic image caption generation

Technology Taxonomy Mapping:
(NASA's technology taxonomy has been developed by the SBIR-STTR program to disseminate awareness of proposed and awarded R/R&D in the agency. It is a listing of over 100 technologies, sorted into broad categories, of interest to NASA.) Autonomous Control (see also Control & Monitoring) Circuits (including ICs; for specific applications, see e.g., Communications, Networking & Signal Transport; Control & Monitoring, Sensors) Computer System Architectures Data Processing Image Analysis Image Processing Intelligence Perception/Vision Robotics (see also Control & Monitoring; Sensors)

Artificial Intelligence realized through Machine Learning algorithms seems to be the only viable solution to implement perception, enable pilot assistants and eventually full autonomy to UAS. Currently, many UAS have some kind of conventional Computer Vision (CV) helping them in obstacle avoidance or target acquisition. Interestingly though, since 2012 Deep Neural Networks (DNN) have dramatically outperformed conventional CV algorithms in those tasks and pushed Artificial Intelligence (AI) limits in a variety of other applications including, but not limited, to object recognition, video analytics, decision making and control, speech recognition, etc. Unfortunately, the computational power required for real-time DNN operation can still only be delivered by bulky, expensive, slow, heavy and energy-hungry digital systems like GPUs.This is why Mentium is devoted to delivering disruptive technology in the field of Machine Learning hardware accelerators, and in particular for this project, into the Deep Learning Hardware Accelerators field. Experimental data and Phase I results confirm that our hardware can deliver 100x to 1000x gain in speed and in power efficiency compared to other state-of-the-art accelerators. Our final product will be able to analyze, in real-time, big data streams coming from cameras, sensors and/or avionics and to categorize (classify) them for the purpose of decision making or object localization to achieve better navigation and collision avoidance in UAS. The same hardware processor will be deployable in the Air Traffic Systems, for real-time data analysis and decision-making. All with more than 10x reduction in cost and power consumption. This distruptive technology is based on an analog-computational core, exploiting the memory devices to carry out the computation at a physical level. Analog computation is inherently faster and more efficient than the digital one, while the in-memory computation removes the data transfer bottleneck.

Potential NASA Commercial Applications:
(Limit 1500 characters, approximately 150 words) We can easily envision our Deep Learning Accelerator deployed in the following NASA's Candidate Mission Products from Thrust 6 Roadmap:-Autonomy-Enabled UAS for Earth Science -Autonomy-Enabled Air Traffic Management-Autonomy-Enhanced Vehicle Safety-Inflight Vehicle Performance optimization-Complex Decision-Making UASWe want also emphasize that the final applications scope is wider than the Aeronautics Directorate, since the same architecture can be optimized for radioactive environments and deployed in space.



Potential NON-NASA Commercial Applications:
:
(Limit 1500 characters, approximately 150 words) Our final product will be attractive both for the private sector and the federal market as well. As for the private sector, the Outcome if this SBIR project can be directly injected into the commercial UAV market, both for consumer and for executive applications.Nevertheless, it is important to consider that with very few modifications our hardware accelerator could be used in:- Industry intelligence (through visual, audio and sensors inputs)- Cybersecurity- Enterprises Big Data analytics- Security cameras- AutomotiveMoreover, number of other federal agencies could be interested in our product, here is a list of the majors:-EPA, Environmental Protection Agencies-USDA, Department of Agriculture-DHS, Department of Homeland Security-DoD, Department of Defense-NOAA, National Oceanic and Atmospheric Administration-DoE, Department of Energy-DoT, Department of Transportation

Technology Taxonomy Mapping:
(NASA's technology taxonomy has been developed by the SBIR-STTR program to disseminate awareness of proposed and awarded R/R&D in the agency. It is a listing of over 100 technologies, sorted into broad categories, of interest to NASA.) Algorithms/Control Software & Systems (see also Autonomous Systems) Autonomous Control (see also Control & Monitoring) Avionics (see also Control and Monitoring) Circuits (including ICs; for specific applications, see e.g., Communications, Networking & Signal Transport; Control & Monitoring, Sensors) Data Processing Image Analysis Intelligence Man-Machine Interaction Perception/Vision Recovery (see also Autonomous Systems) ---------- By a direct extension of our Phase II contract with NASA, we propose to realize a prototype of a neuromorphic chip able to withstand heavy dosage of radiations, like those encountered in the Lunar environment. The neuromorphic chip will have multiple use in a space system, from sensor data analysis to control or to system diagnosis.In particular, during our concluded Phase II we realized a test chip in silicon that has been tested on the ground. That test chip demonstrated the architecture, the computing capabilities and clearly showed the potential to reach unprecedented speed and efficiency.The architecture revolves around a analog in-memory computing architecture, a non-von Neumann design that unifies the computing module and the memory module, which becomes one and the same, removing at its root the memory transfer bottleneck, hybridized with a co-designed digital accelerator. This hybrid architecture offers incredible advantages over purely digital architectures.A key part of the innovation will reside in the use of memristors or ReRAMs. We will team up with a tier 1 company in the semiconductor fabrication processes: Applied Materials.nbsp;The physics at the base of the memristor non-volatile memory behavior makes them particularly resilient to radiations.nbsp;In summary, in 20nbsp;months we will combine our experience in in-memory computing architectures built during the Phase II project with the Applied Materials processing experience to deliver a TRL 7 integrated system,nbsp;with a hybrid of digital and memristor-based in-memory computing chips, with the potential to unleash an unprecedented level of AI capabilities for Lunar explorations and beyond.