“Real-time and adaptive learning is the stepping stone for the adoption of AI in our everyday lives at home, work, public places, and more,” says Guy Paillet, President, and CEO of General Vision.
SINGAPORE – Media OutReach – June 6, 2017 – NeuroMem Technologies Pte Ltd, A General Vision Inc. Company, focused on the use and benefits of neuromorphic semiconductor components for pattern recognition announced that its Neuromorphic Memory Unit (NMU) IP has been structured for ease of implementation into semiconductor designs and is available for immediate licensing. NeuroMem’s massively parallel solution enables pattern recognition at a fraction of the time and power used by conventional processing techniques. Our NMU technology has already been licensed by top 5 semiconductor manufacturers for incorporation into their chips, and the newly generated IP package has been validated, and developed to ease adoption for future designs.
Pattern matching is everywhere. As applications in IoT, context awareness and autonomous vehicles begin to see broader adoption the need for fast, efficient, low power matching is becoming critical.
Current pattern matching techniques, based on traditional processing cores like DSPs, CPUs, and GPUs take the pattern and search, one at a time, through a list of possible patterns. This presents a fundamental problem when you have to search across a large database of potential matches. The time and power it will take to make decisions using conventional approaches are quickly becoming unpractical.
“Today we are seeing many people realize that traditional techniques will be inadequate for dealing with the tsunami of data and speed at which it needs to be processed, said Pierre Brunswick, Chairman of NeuroMem Technologies. A General Vision Inc. Company and these innovators understand that neuromorphic solutions from Singapore will take us to next age of artificial intelligence.”
The fundamental core of the technology is based on massively parallel pattern matching capability. The time it takes to match a pattern to a single trained neuron can be the same is it would take to match against a million neurons, currently a fraction of a microsecond. Our technology includes the ability to realize that a pattern has not been trained and represents novelty, which can be learned on the fly, while also continuing to operate. This automatic generation of models is done without the need to write any software to support the learning. The techniques are based on how the brain learns by associating stimuli to an output which can be a name, an action or experience. As a child you saw, heard, or felt something, and someone told you what that thing was. You stored the pattern in your memory and the next time you encountered it you recalled what it was. The true benefit of learning by examples is that anyone can transfer his knowledge by creating a training set, from a person working on a factory floor to assure top-notch quality products, to someone adding deep layers of personalization on their phone or smartwatch.
NeuroMem Technologies, A General Vision Inc. Company, is licensing NeuroMem IP, which will allow everyday objects to have perception of their environment and interact with users. We truly believe that pattern learning and recognition can become practical and ubiquitous only, if it can rely on components inspired by the human brain, which we call neuromorphic memories (1) merging storage and local processing per cell, and (2) massively parallel interconnected cells operating at low power.