Securing the path to the memristive neural processor (MNP) KnowmTech was formed in response to a request for consulting services in neuromorphic computing for DARPA's SyNAPSE and Physical Intelligence program., KnowmTech, LLC is focused to developing breakthrough machine learning technologies that address fundamental problems inherent in modern computing architectures. The separation of memory and processing in computing architectures leads to wasted energy shuttling information back and forth - causing problems for simulations of adaptive operations-exactly the operations needed for things like machine learning (ML). Since ML is inherently about adaptation, it turns out that building large-scale adaptive systems like brains - at a power efficiency close to biology - is simply not feasible with traditional computing approaches. Current super-computing installations consume the equivalent of a "Three-Gorges Dam's worth"-over 10 Gwatts-to emulate one human-scale brain at low resolution in real-time. AHaH computing changes this by taking advantage of a universal adaptive building block found throughout nature and mapping this to electronic circuits. The physical circuit turns out to be exceedingly simple, allowing for various levels of simulation abstraction - including a very efficient digital emulator that runs on modern hardware. A first prototype of a KnowmTM-based neural processing unit (NPU) called Thermodynamic RAM is currently being pursued with development partners this summer. Buildingfrom the bottom up, KnowmTech started with nano-particles, went to memristors, then synapses and neurons. Now we have machine learning modules across the domains of perception, planning and control.
