A Systematic Method for Configuring VLSI Networks of Spiking Neurons

Abstract

An increasing number of research groups are developing custom hybrid analog/digital very large scale integration (VLSI) chips and systems that implement hundreds to thousands of spiking neurons with biophysically realistic dynamics, with the intention of emulating brainlike real-world behavior in hardware and robotic systems rather than simply simulating their performance on general-purpose digital computers. Although the electronic engineering aspects of these emulation systems is proceeding well, progress toward the actual emulation of brainlike tasks is restricted by the lack of suitable high-level configuration methods of the kind that have already been developed over many decades for simulations on general-purpose computers. The key difficulty is that the dynamics of the CMOS electronic analogs are determined by transistor biases that do not map simply to the parameter types and values used in typical abstract mathematical models of neurons and their networks. Here we provide a general method for resolving this difficulty. We describe a parameter mapping technique that permits an automatic configuration of VLSI neural networks so that their electronic emulation conforms to a higher-level neuronal simulation. We show that the neurons configured by our method exhibit spike timing statistics and temporal dynamics that are the same as those observed in the software simulated neurons and, in particular, that the key parameters of recurrent VLSI neural networks (e.g., implementing soft winner-take-all) can be precisely tuned. The proposed method permits a seamless integration between software simulations with hardware emulations and intertranslatability between the parameters of abstract neuronal models and their emulation counterparts. Most important, our method offers a route toward a high-level task configuration language for neuromorphic VLSI systems. An increasing number of research groups are developing custom hybrid analog/digital very large scale integration (VLSI) chips and systems that implement hundreds to thousands of spiking neurons with biophysically realistic dynamics, with the intention of emulating brainlike real-world behavior in hardware and robotic systems rather than simply simulating their performance on general-purpose digital computers. Although the electronic engineering aspects of these emulation systems is proceeding well, progress toward the actual emulation of brainlike tasks is restricted by the lack of suitable high-level configuration methods of the kind that have already been developed over many decades for simulations on general-purpose computers. The key difficulty is that the dynamics of the CMOS electronic analogs are determined by transistor biases that do not map simply to the parameter types and values used in typical abstract mathematical models of neurons and their networks. Here we provide a general method for resolving this difficulty. We describe a parameter mapping technique that permits an automatic configuration of VLSI neural networks so that their electr...

Synthesizing cognition in neuromorphic electronic systems

“…Second, the abstract computational layer, composed of modular sWTA networks, can be easily mapped onto multiple neuromorphic chips. Third, the parameters of the sWTA networks can be reliably and automatically mapped onto the electronic voltage and current biases of the CMOS electronic neurons (19). Finally, only a small number of connections between the transition neurons and the state populations need to be specified or learned to achieve a desired functionality.…”

“…Slow time constants in the network result in a "low noise regime", in which the mathematical formulation of the dynamics becomes tractable. In this regime, the firing rate of a spiking neuron population can be approximated with a threshold-linear activation function, σð·Þ ¼ maxð·; 0Þ (19,63). For convenience, we call a population described by a mean-field state variable a linear threshold unit (LTU).…”

“…The state variable x describes the "synaptic state", instead of the firing rates, but these two are proportional in steady state (19). In this interpretation, τ i is equal to the synaptic time constant.…”

“…The parameters of the sWTA networks were initially configured using the methods described in ref. 19. A calibration mechanism described in ref.…”

“…We solve this problem by an automatic method that involves using a mean-field approach to measuring the steady-state firing rate response of the VLSI neurons, deriving the average neural intracellular current that generated this activity, and then fitting the parameters of the CMOS transistor model to provide the estimated neuronal current-discharge relation. In this way the method generates the necessary mappings between the parameters of the sWTA network models and their electronic counterparts (19).…”