ABSTRACT
We present a novel one-transistor/one-resistor (1T1R) synapse for neuromorphic networks, based on phase change memory (PCM) technology. The synapse is capable of spike-timing dependent plasticity (STDP), where gradual potentiation relies on set transition, namely crystallization, in the PCM, while depression is achieved via reset or amorphization of a chalcogenide active volume. STDP characteristics are demonstrated by experiments under variable initial conditions and number of pulses. Finally, we support the applicability of the 1T1R synapse for learning and recognition of visual patterns by simulations of fully connected neuromorphic networks with 2 or 3 layers with high recognition efficiency. The proposed scheme provides a feasible low-power solution for on-line unsupervised machine learning in smart reconfigurable sensors.
ABSTRACT
Today phase change memory (PCM) technology has reached product maturity at 90 and 65 nm nodes, while the 45 nm node is under development and is expected to enter in the market soon. The continuous decrease of the cell size with scaling leads to an effective active area as small as 150 nm(2) and an active volume involved in the phase transformation of about 10(4) nm(3), thus entering definitively into the nanotechnology world. At this extremely reduced dimension, the reliability of the device must be carefully investigated. In this work we show that the cycling performance of the device is well maintained, not being a problem for either the bipolar transistor or the storage element. The phase transition from the amorphous to the crystalline state is, of course, one of the most interesting phenomena, impacting cell retention capability and device performance. The stochastic nature of nano-nuclei percolation in the amorphous matrix is shown as an important ingredient in the retention of PCM devices. The related dispersion in crystallization times is analyzed through a crystallization Monte Carlo model and a physical insight into nucleation and growth mechanisms is provided.