EEtimes – researchers are finding phase-change memory (PCM) technology a good candidate for use as programmable artificial synapses in neuromorphic circuitry because of its great energy efficiency, scaling potential, endurance and reliability.
Researchers from a combination of several European labs led by CEA, Leti and Minatec will report on their search to reduce the high power consumption of large neuromorphic circuits which can be used for complex tasks such as pattern-recognition.
After investigating various phase-change memory devices for use as programmable synapses in an ultra-dense, large-scale neuromorphic system the researchers designed, modeled and simulated a neural network based on a two-PCM-device per synapse structure (four million PCMs in total)
heir work indicates such a large system can recognize complex patterns with 92-percent accuracy, and with system power consumption of just 112µW in “learning” mode.
Meanwhile, Stanford University researchers will report on experiments they conducted and computer simulations they performed to investigate the use of programmable PCM synapses in brain-inspired electronic systems that could implement STDP.
STDP–spike-timing-dependent plasticity–is tied to synaptic plasticity, a fundamental brain mechanism for learning and memory. Electronic analogs which mimic the relative timing of these neuronal spikes have been used in computer studies of pattern-recognition, directionality, navigation, time-sequence learning and coincidence-detection.
The Stanford researchers configured a crossbar array of 75-nm PCM devices made from Germanium-Antimony-Tellurium (GeSbTe), a phase change chalcogenide material, The devices featured picojoule-level power consumption, which for a system with 10^10 synapses should lead to total power consumption of only 10 W for all synaptic activity, according to the researchers.
(Paper #4.4, “Phase Change Memory as Synapse for Ultra-Dense Neuromorphic Systems: Application to Complex Visual Pattern Recognition,” M. Suri et al, CEA-LETI-MINATEC/CEA-LIST, LCE/CNRS-IEMN)
(Paper #30.3, “Energy-Efficient Programming of Nanoelectronic Synaptic Devices for Large-Scale Implementation of Associative and Temporal Sequence Learning,” D. Kuzum et al, Stanford University)