Implementing homeostatic plasticity in VLSI networks of spiking neurons

Bartolozzi, Chiara; Nikolayeva, O.; Indiveri, Giacomo

doi:10.1109/icecs.2008.4674945

Cited by 27 publications

(18 citation statements)

References 16 publications

(21 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is more challenging for analog neuromorphic systems, since homeostasis requires particularly long time constants. Both pure analog and mixed analog/digital solutions have been proposed [89]. In terms of inference, homeostasis implements a prior belief that all latent variable values have a significant chance of occurring.…”

Section: Robustness To Other Issuesmentioning

confidence: 99%

Bioinspired Programming of Memory Devices for Implementing an Inference Engine

Querlioz¹,

Bichler²,

Vincent³

et al. 2015

Proc. IEEE

121

View full text Add to dashboard Cite

Emerging memory structures have several characteristics that are suitable for neuromorphic implementations. This paper connects the behavior of these ''new'' memory devices to achieving inference engines and provides a connection between the behavior of different devices and the learning algorithms.ABSTRACT | Cognitive tasks are essential for the modern applications of electronics, and rely on the capability to perform inference. The Von Neumann bottleneck is an important issue for such tasks, and emerging memory devices offer an opportunity to overcome this issue by fusing computing and memory, in nonvolatile instant ON/OFF systems. A vision for accomplishing this is to use brain-inspired architectures, which excel at inference and do not differentiate between computing and memory. In this work, we use a neuroscience-inspired model of learning, spike-timing-dependent plasticity, to develop a bioinspired approach for programming memory devices, which naturally gives rise to an inference engine. The method is then adapted to different memory devices, including multivalued memories (cumulative memristive device, phasechange memory) and stochastic binary memories (conductive bridge memory, spin transfer torque magnetic tunnel junction). By means of system-level simulations, we investigate several applications, including image recognition and pattern detection within video and auditory data. We compare the results of the different devices. Stochastic binary devices require the use of redundancy, the extent of which depends tremendously on the considered task. A theoretical analysis allows us to understand how the various devices differ, and ties the inference engine to the machine learning algorithm of expectation-maximization. Monte Carlo simulations demonstrate an exceptional robustness of the inference engine with respect to device variations and other issues. A theoretical analysis explains the roots of this robustness. These results highlight a possible new bioinspired paradigm for programming emerging memory devices, allowing the natural learning of a complex inference engine. The physics of the memory devices plays an active role. The results open the way for a reinvention of the role of memory, when solving inference tasks.

show abstract

Section: Robustness To Other Issuesmentioning

confidence: 99%

Bioinspired Programming of Memory Devices for Implementing an Inference Engine

Querlioz¹,

Bichler²,

Vincent³

et al. 2015

Proc. IEEE

121

View full text Add to dashboard Cite

show abstract

“…In the future, we hope to expand the circuit's operating range both in time and in controllability. While previous work has demonstrated time constants on the order of hundreds of seconds using floating gates [4], our simulations have only demonstrated time constants in range of several seconds (similar to [6]). A potential solution is to decrease the injection voltage in the adaptation circuitry, but this may affect other circuit parameters such as the range of allowable input currents.…”

Section: Simulation Resultsmentioning

confidence: 60%

“…In contrast, our work features an explicit circuit and associated bias voltage which controls the steady-state response. Homeostasis was implemented in the form of synaptic scaling by manipulating a free parameter in the Differential Pair Integrator synapse [5] with both a custom analog controller [6] and an offline digital controller [7]. The digital controller showed that homeostatic plasticity can indeed be achieved with a Proportional-Integral-Derivative (PID) controller mechanism, but the use of an offline control mechanism is undesirable for real applications with large numbers of neurons.…”

Section: Introductionmentioning

confidence: 99%

Floating-gate-based intrinsic plasticity with low-voltage rate control

Nease

Chicca

2016

2016 IEEE International Symposium on Circuits and Systems (ISCAS)

View full text Add to dashboard Cite

Homeostatic intrinsic plasticity is a neurobiological mechanism which adapts a neuron's channel properties over long timescales to achieve a desired average firing rate. This mechanism works to stabilize neural network dynamics. We present simulation results of a neuromorphic circuit implementation of homeostatic intrinsic plasticity using floating-gate transistors. While high voltages are required for tunneling and injection, the control parameter in our implementation is a low voltage, allowing for easy interfacing with typical analog CMOS control circuitry.

show abstract

“…However, despite being extremely important for the design of large scale neuromorphic computing platforms, only few works addressed the implementation of homeostasis in silicon neural networks, mainly because of the technical difficulty in obtaining the necessary extremely long time constants with the intrinsically fast CMOS circuital elements. Existing approaches focus on the use of floating gate transistors [5], [6], or propose to use off-chip methods, that require external memory and digital circuits [7]. Here we propose a novel auto-gain synaptic scaling circuit that exploits the features of an ultra-low leakage cell implemented using standard CMOS technology able to achieve extremely long time constant [8]- [10].…”

Section: Imentioning

confidence: 99%

Automatic gain control of ultra-low leakage synaptic scaling homeostatic plasticity circuits

Qiao

Indiveri

Bartolozzi

2016

2016 IEEE Biomedical Circuits and Systems Conference (BioCAS)

Self Cite

View full text Add to dashboard Cite

Homeostatic plasticity is a stabilizing mechanism that allows neural systems to maintain their activity around a functional operating point. This is an extremely useful mechanism for neuromorphic computing systems, as it can be used to compensate for chronic shifts, for example due to changes in the network structure. However, it is important that this plasticity mechanism operates on time scales that are much longer than conventional synaptic plasticity ones, in order to not interfere with the learning process. In this paper we present a novel ultra-low leakage cell and an automatic gain control scheme that can adapt the gain of analog log-domain synapse circuits over extremely long time scales. To validate the proposed scheme, we implemented the ultra-low leakage cell in a standard 180 nm Complementary Metal-Oxide-Semiconductor (CMOS) process, and integrated it in an array of dynamic synapses connected to an adaptive integrate and fire neuron. We describe the circuit and demonstrate how it can be configured to scale the gain of all synapses afferent to the silicon neuron in a way to keep the neuron's average firing rate constant around a set operating point. The circuit occupies a silicon area of 84 µm×22 µm and consumes approximately 10.8 nW with a 1.8 V supply voltage. It exhibits time constants of up to 25 kilo-seconds, thanks to a controllable leakage current that can be scaled down to 1.2 atto-Amps (7.5 electrons/s).

show abstract

Implementing homeostatic plasticity in VLSI networks of spiking neurons

Cited by 27 publications

References 16 publications

Bioinspired Programming of Memory Devices for Implementing an Inference Engine

Bioinspired Programming of Memory Devices for Implementing an Inference Engine

Floating-gate-based intrinsic plasticity with low-voltage rate control

Automatic gain control of ultra-low leakage synaptic scaling homeostatic plasticity circuits

Contact Info

Product

Resources

About