Design of a modular and mixed neuromimetic ASIC

Abstract

This paper presents a new specific integrated circuit (ASIC) that emulates neurons electrical activity using a biophysical model (neuromimetic ASIC). Such ASICs form the computation core of a complete simulation system dedicated to the investigation of the dynamics of biomimetic neural networks. The circuits were designed using a modular approach. Simulations were realized by mixing the behavioral and the transistor-level description of modules. We present in this paper the ASICs specifications and architecture, together with simulation results.

A Library of Analog Operators Based on the Hodgkin-Huxley Formalism for the Design of Tunable, Real-Time, Silicon Neurons

“…We have already used this library to design another IC dedicated to the real-time hardware simulation of medium-sized neural networks (up to 128 neurons). This IC includes 5 analog cores, synapses, and digital functions to manage the analog components [51]. We are also working on the design of an embedded simulation platform merging analog conductance-based silicon neurons and digital adaptive synapses [52].…”

Cost-efficient FPGA implementation of basal ganglia and their Parkinsonian analysis

“…very large scale integration (VLSI), special purpose hardware (SPH), and graphical processing unit (GPU)-based and field programmable gate array (FPGA)-based methods for simulating biological dynamics. VLSI designs can be more realistic because of the capacity to implement nonlinear models in a straightforward manner (Fieres, Schemmel, & Meier, 2008;Lewis & Renaud, 2007;Mahowald & Douglas, 1991;Renaud, Tomas, Bornat, Daouzli, & Saighi, 2007;Tomas, Bornat, Saighi, Levi, & Renaud, 2006), but the high costs and long development time of this approach limit its applications. SPH-based design has been employed to implement neurobiological functions (Furber et al, 2013;Furber, Temple, & Brown, 2006), but it is also affected by high costs and long development time.…”

Application of IP-Based Analog Platforms in the Design of Neuromimetic Integrated Circuits

“…Our work was guided by these concepts and developments, and also took into account the specificities of neuromimetics circuits and systems. Our background material results from a 20 years experience in our research group in this field of neuromorphic engineering [4] [5]. It consists in particular in a library of analog computation blocks used to design analog ASICs (Application Specific Integrated Circuit) that mimic the dynamics of biological neurons electrical activity [6].…”

“…Then, we define our methodology in 3 points: the IP-package content (section IV), the design of a re-usable IP block (section V) and the integration of an IP in the system design flow (section VI). Concrete application of the methodology is given in section VII, through examples of design projects [4]. II.…”