Hi Esq,
I have a file on this from a month ago.
Researchers from IBM’s Zurich lab have argued that these recognition techniques could be improved by enhancing AI hardware. The goal is to use the well-known phase-change memory (PCM) technology to develop a new type of artificial synapse. The researchers employ a PCM memtransistive synapse, which combines memristors, a nonvolatile electronic memory element, and transistors into a single low-power device. This shows a non-Von Neumann in-memory computing architecture that offers various powerful cognitive frameworks for ML applications, such as short-term spike-timing-dependent plasticity and probabilistic Hopfield Neural Networks.
In-memory computing has become popular in the last year or two - an attempt to get around the von Neumann bottleneck on the bus from the memory to the processor as the processor attempts to retrieve data from the memory. Akida is similar in that the memory is distributed amongst its 80 NPUs (Neuromorphic Processing Units).
A phase change material is one whose resistance can be changed by applying a large voltage in one direction, and reversed by applying the voltage in the opposite direction. The change in resistance can be detected by sensing the size of the current when a lower voltage is applied.
IBM has been playing around with phase change-materials for over a decade.
US2010223220A1 ELECTRONIC SYNAPSE
View attachment 6505
They are familiar with STDP:
WO2010133399A1 ELECTRONIC LEARNING SYNAPSE WITH SPIKE-TIMING DEPENDENT PLASTICITY USING PHASE CHANGE MEMORY
View attachment 6506
They also claim unsupervised learning:
US11164080B2 Unsupervised, supervised and reinforced learning via spiking computation
View attachment 6507
Even when IBM did summon up the courage to dip their toes in the digital neuron pond, they still kept a lifeline firmly attached to the analog neuron anchor:
WO2014080300A1 NEURAL NETWORK
"0016]
The term digital neuron as used herein represents an framework configured to simulate a biological neuron. An digital neuron creates connections between processing elements that are roughly functionally equivalent to neurons of a biological brain. As such, a neuromorphic and synaptronic computation comprising digital neurons, according to embodiments of the invention, may include various electronic circuits that are modeled on biological neurons. Further, a neuromorphic and synaptronic computation comprising digital neurons, according to embodiments of the invention, may include various processing elements (including computer simulations) that are modeled on biological neurons.
Although certain illustrative embodiments of the invention are described herein using digital neurons comprising digital circuits, the present invention is not limited to digital circuits. A neuromorphic and synaptronic computation, according to embodiments of the invention, can be implemented as a neuromorphic and synaptronic framework comprising circuitry and additionally as a computer simulation. Indeed, embodiments of the invention can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment containing both hardware and software elements."
Like most of the NNs that I've seen coming from Switzerland, with one exception, they all use analog neurons/synapses. The analog neuron is a closer replica of the human neuron, but, in silicon, it is difficult to reproduce the neurons with sufficient accuracy, so that the threshold voltage for the neurons with "identical" inputs can differ between neurons. Analog neurons add the voltage or current of the spikes, but the amplitude of the voltage/current is not consistent between neurons because of manufacturing variations. Given that neurons can have hundreds of inputs, the errors accumulate to give inconsistent performance with
analog neurons. With Akida's
digital neurons this is not a problem because the digital bits are ON or OFF (1 or zero), and it was PvdM's recognition of this that has put us where we are today.
www.arm.com
