212 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 2, NO. 3, SEPTEMBER 2008 A Silicon Central Pattern Generator Controls Locomotion in Vivo R. Jacob Vogelstein, Member, IEEE, Francesco V. G. Tenore, Member, IEEE, Lisa Guevremont, Member, IEEE, Ralph Etienne-Cummings, Senior Member, IEEE, and Vivian K. Mushahwar, Member, IEEE Abstract—We present a neuromorphic silicon chip that emu- lates the activity of the biological spinal central pattern generator (CPG) and creates locomotor patterns to support walking. The chip implements ten integrate-and-fire silicon neurons and 190 programmable digital-to-analog converters that act as synapses. This architecture allows for each neuron to make synaptic connec- tions to any of the other neurons as well as to any of eight external input signals and one tonic bias input. The chip’s functionality is confirmed by a series of experiments in which it controls the motor output of a paralyzed animal in real-time and enables it to walk along a three-meter platform. The walking is controlled under closed-loop conditions with the aide of sensory feedback that is recorded from the animal’s legs and fed into the silicon CPG. Although we and others have previously described biomimetic silicon locomotor control systems for robots, this is the first demonstration of a neuromorphic device that can replace some functions of the central nervous system in vivo. Index Terms—Central pattern generator (CPG), neuromorphic, neuroprosthesis, silicon, VLSI. I. INTRODUCTION N EUROPROSTHETIC devices aim to replace non- functional parts of the nervous system with artificial components and restore functions lost due to injury or disease. Current neuroprostheses (e.g., cochlear implants) typically require large external modules and frequent recharging due to power-hungry digital processors. Although neuromorphic Manuscript received May 21, 2008; revised March 16, 2008. Current ver- sion published October 24, 2008. This work was supported in part by the Na- tional Science Foundation (NSF) ERC Cooperative Agreement #EEC9731478, the Office of Naval Research, the National Institute of Neurological Disorders and Stroke, the Alberta Heritage Foundation for Medical Research, the Interna- tional Spinal Research Trust, and an NSF Graduate Research Fellowship to R. Vogelstein. R. Vogelstein and F. Tenorecontributed equally to this work. The authors would like to thank the NSF-sponsored Telluride Neuromorphic Engi- neering Workshop for providing a forum for this collaboration. This paper was recommended by Associate Editor M. Sawan. R. J. Vogelstein is with the Department of Biomedical Engineering, Johns Hopkins University, Baltimore, MD 21218 USA (e-mail: jvogelst@bme.jhu. edu). F. V. G. Tenore and R. Etienne-Cummings are with the Department of Elec- trical and Computer Engineering, Johns Hopkins University, Baltimore, MD 21218 USA (e-mail: fra@jhu.edu, retienne@jhu.edu). L. Guevremont is with the Department of Biomedical Engineering and the Center for Neuroscience, University of Alberta, Edmonton, AB, T6G 2S2 Canada (e-mail: lg@ualberta.ca). V. K. Mushahwar is with the Department of Cell Biology and the Center for Neuroscience, University of Alberta, Edmonton, AB, T6G 2S2 Canada (e-mail: vivian.mushahwar@ualberta.ca). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TBCAS.2008.2001867 processors [1] do not presently contain equivalent processing power to their digital counterparts, they have been proposed as an alternative solution for future neuroprostheses because by emulating the structure and function of biological neurons (and with the appropriate design constraints), they can be used in compact, low-power, and implantable devices [2]–[4]. Here we present a novel neuromorphic silicon central pattern generator (SiCPG) chip and demonstrate its use as a primitive neuroprosthetic replacement for some functions of the spinal CPG for locomotion. Central pattern generators are specialized neural circuits that produce rhythmic, patterned outputs to control motor systems. In many mammals, the CPG that controls hind limb locomotion is located in the lumbar spinal cord [5]. Under normal condi- tions, the CPG continuously integrates descending and afferent inputs to ensure production of smooth motor trajectories [6]. After a severe spinal cord injury, descending fiber tracts are dis- rupted and the brain cannot initiate activity in the CPG, resulting in paralysis. Although we and others have previously suggested that a neuromorphic CPG could serve as an in vivo replacement for the natural CPG after spinal cord injury [7], this had not pre- viously been demonstrated. Within the past ten years, a number of neuromorphic SiCPG chips have been presented in the literature with different num- bers of on-chip neurons and varying degrees of biological re- alism [8]–[16]. Simoni et al. designed a chip around a very detailed multi-conductance Hodgkin–Huxley model [17], but implemented only a single neuron on a 4 mm die [11]. Cym- balyuk et al., Lewis et al., and Nakada et al. have each cre- ated chips with two oscillating silicon neurons (based on the Morris–Lecar model [8], [18], a simple integrate-and-fire model [9], and the Matsuoka model [12], [19], respectively). Kier et al. and Nakada et al. both built four-neuron chips, the former based on a modified Hopfield model [14], [20] and the latter based on the Amari–Hopfield neural model [10], [21]. Finally, Arena et al. have demonstrated a six-neuron CPG chip based on a cel- lular neural network architecture [13]. All of these chips contain some silicon analogue of biological neurons and allow for some degree of programmable connectivity between cells. When the different types of neurons are configured in appropriate network architectures, they can form oscillatory units that resemble their biological counterparts [22]. Just like the biological CPG, how- ever, a SiCPG must also respond to sensory feedback from the limbs to reliably produce locomotion in vivo. In the following sections, we first present the design of a ten-neuron reprogrammable neuromorphic SiCPG chip that can utilize external sensory information to modulate the activity of its silicon neurons [15]. We then show results from experiments 1932-4545/$25.00 © 2008 IEEE