Addition of HfO2 interface layer for improved synaptic performance of phase change memory (PCM) devices

• Published in: Solid-state electronics Vol. 79; pp. 227 - 232
• Main Authors: Suri, M., Bichler, O., Hubert, Q., Perniola, L., Sousa, V., Jahan, C., Vuillaume, D., Gamrat, C., DeSalvo, B.
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 01.01.2013; Elsevier
• Subjects: Applied sciences; Automobiles; Automotive engineering; Bio-inspired computing; Cars; Computer simulation; Design. Technologies. Operation analysis. Testing; Devices; Electric, optical and optoelectronic circuits; Electronics; Exact sciences and technology; Extraction; Hafnium oxide; Integrated circuits; Integrated circuits by function (including memories and processors); Interface engineering; Neural networks; Phase change; Phase change memory; Semiconductor electronics. Microelectronics. Optoelectronics. Solid state devices; Spike time dependent plasticity; Visual pattern extraction; Interface engineering; Phase change memory; Bio-inspired computing; Visual pattern extraction; Spike time dependent plasticity; Interfacial layer; Performance evaluation; System architecture; Spiking neuron; Random access memory; Hafnium oxide; Shape detection; Thin film; Learning; Time dependence; Biomimetics; Non volatile memory; Energetic efficiency; Plasticity; PCM material; Pattern extraction; Large scale system; Neural network; Synapse; Integrated circuit
• ISSN: 0038-1101; 1879-2405
• DOI: 10.1016/j.sse.2012.09.006

► Methodology to use PCM synapses for neuromorphic visual pattern extraction. ► Programming power reduction of PCM synapses through addition of HfO2 interface. ► Improved performance by increasing the number of intermediate resistance states. ► Greater than 50% reduction in system power dissipated in the learning mode. In this work, we will focus on the use of phase change memory (PCM) to emulate synaptic behavior in emerging neuromorphic system-architectures. In particular, we will show that the performance and energy-efficiency of large scale neuromorphic systems can be improved by engineering individual PCM devices used as synapses. This is obtained by adding a thin HfO2 interface layer to standard GST PCM devices, allowing for the lowering of the Set/Reset currents and the increase of the number of intermediate resistance states (or synaptic weights) in the synaptic potentiation characteristics. The experimentally obtained potentiation characteristics of such PCM devices are used to simulate a 2-layer ultra-dense spiking neural network (SNN) and to perform a complex visual pattern extraction from a test case based on real world data (i.e. cars passing on a 6-lane freeway). The total power dissipated in the learning mode, for the pattern extraction experiment is estimated to be as low as 60μW. Average detection rate of cars is found to be greater than 90%.