Hybrid CMOS-RRAM Neuron circuits
Publié le : 8 octobre 2019
Brain-inspired architectures in neuromorphic hardware are currently subject to intensive research as an alternative to the limits of traditional computer organization. The remarkable computing performance and efficiency of biological nervous systems are widely attributed to the co-localization of memory and computation spatially through the structure. Re-configurable non-volatile resistive memories (RRAMs) can be incorporated into neuron and synapse circuit models allowing memory to be truly co-localized with the computational units in the computing fabric facilitating the realization of massively parallel local plasticity mechanisms in neuromorphic hardware. Hybrid CMOS-RRAM Neurons have been recently proposed by Leti. RRAM memories allow to locally store the neuron parameters, which is a fundamental precondition for adapting the computation to the scale of input signals through the implementation of neuronal intrinsic plasticity.
Expected work :
The proposed internship gives the opportunity to challenge new RRAM applications. Simple hybrid CMOS-RRAM neuron circuits storing neuron parameters within RRAM will be available for the beginning of the internship position. The main goal of the proposed internship is the extensive experimental study of these circuits. The impact of the RRAM programming conditions (voltage/time) and power consumption on the circuit performances will be addressed. The second objective will be to elaborate and test new strategies to implement neuronal intrinsic plasticity, where neuron adapts its properties (the parameters stored in the RRAM memories) to maximize its information capacity based on the statistical properties of its input while minimizing the power it consumes.
If you are interested by the internship, please send your CV and a motivation letter to email@example.com