A biophysically-based neuromorphic model of spike rate- and timing-dependent plasticity

G. Rachmuth, H. Z. Shouval, M. F. Bear, C.-S. Poon
2011 Proceedings of the National Academy of Sciences of the United States of America  
Learning and memory are emergent animal behaviors governed by modifications in neural activity in response to changing environments. Modification of neural activity is driven partly by changing the connection strength between neurons in a process called synaptic plasticity-the degree to which a presynaptic neuron can trigger a postsynaptic one to produce a nerve impulse-following the paired activation of the preand postsynaptic neurons. Two classic paradigms for inducing Hebbian synaptic
more » ... ity (1) in the mammalian hippocampus (the region of the brain associated with memory formation) and the neocortex (the gray matter) are spike rate-dependent plasticity (SRDP), in which the sign and magnitude of synaptic plasticity are determined by presynaptic firing rate; and spike-timingdependent plasticity (STDP), in which the precise timing of pre-and postsynaptic activities determines the direction and strength of synaptic plasticity. The two protocols differ in their information transfer capabilities, but mechanistically, the induction protocols of SRDP and STDP activate similar calcium-dependent processes that lead to (i) the induction of synaptic long-term potentiation (LTP) (the long-term enhancement of synaptic excitatory strength) or (ii) long-term depression (LTD). This common mechanistic link suggests a possible underlying interrelationship between these two seemingly distinct forms of Hebbian synaptic plasticity (2). Previous modeling studies of SRDP and STDP were mostly based on numerical simulations of model equations on digital computers. Compared to digital computers, neuromorphic (or neuro-inspired) electronic circuits have an extremely small size and low power requirements for modeling the neural system at a large scale and performing simulations at a high speed. Here, we propose an analog very-large-scale-integrated (VLSI) circuit implementation of "learning synapse" that includes circuit model of an excitatory postsynaptic compartment in a hippocampal neuron, and produces both the SRDP and STDP rules for the induction of LTP and LTD. The chip contains CMOS (complementary metal-oxidesemiconductor) building-block circuits biased in the subthreshold regime for modeling ionic signaling (iono-neuromorphic) as described previously (3) . The circuits allow tremendous flexibility in emulating synapses from various brain structures by simply tuning a small (1-4) set of parameters. The learning-rule implementation underlying on-chip synaptic plasticity is an adaptation of a model proposed by Shouval, et al. (4) that relies on calcium dynamics in the cell to determine synaptic plasticity. The iono-neuromorphic synapse design is biologically intuitive and allows the application of experimental manipulations to observe emergent behaviors. All simulations were in real biological time. We tested the learning synapse circuit by using several well known induction protocols to draw direct comparisons with biological preparations. We showed that SRDP induction protocols reproduce the classical calcium-dependent plasticity. When we subjected the learning synapse to an STDP stimulation protocol, our results reproduced the LTP portion of the STDP window. For LTD, the situation is much more complicated than for LTP. For the artificial synapse circuit with only the NMDA channel as a calcium source, STDP protocols did not display an abrupt transition in the calcium level around Δt ∼ 0 (4). We therefore hypothesized a second coincidence detector may lie beyond the NMDA channel, as discussed in several recent papers. A second biophysical coincidence detector that accounts for postpre LTD may be endogenous cannabinoid (endocannabinoid) molecules (5) (Fig. P1A) . To account for possible coincident detection via retrograde signaling, we Fig. P1. (A) An improved synapse model including presynaptic circuits of NMDA autoreceptors and cannabinoid receptors type 1. (B) Chip results showing the LTD portion of STDP window. Blue line represents an average of several runs.
doi:10.1073/pnas.1106161108 pmid:22089232 pmcid:PMC3241759 fatcat:gut4s6gm6zeejjuewo34vgpz2y