Mapping Generative Models onto a Network of Digital Spiking Neurons.

Stochastic neural networks such as Restricted Boltzmann Machines (RBMs) have been successfully used in applications ranging from speech recognition to image classification, and are particularly interesting because of their potential for generative tasks. Inference and learning in these algorithms use a Markov Chain Monte Carlo procedure called Gibbs sampling, where a logistic function forms the kernel of this sampler. On the other side of the spectrum, neuromorphic systems have shown great promise for low-power and parallelized cognitive computing, but lack well-suited applications and automation procedures. In this work, we propose a systematic method for bridging the RBM algorithm and digital neuromorphic systems, with a generative pattern completion task as proof of concept. For this, we first propose a method of producing the Gibbs sampler using bio-inspired digital noisy integrate-and-fire neurons. Next, we describe the process of mapping generative RBMs trained offline onto the IBM TrueNorth neurosynaptic processor-a low-power digital neuromorphic VLSI substrate. Mapping these algorithms onto neuromorphic hardware presents unique challenges in network connectivity and weight and bias quantization, which, in turn, require architectural and design strategies for the physical realization. Generative performance is analyzed to validate the neuromorphic requirements and to best select the neuron parameters for the model. Lastly, we describe a design automation procedure which achieves optimal resource usage, accounting for the novel hardware adaptations. This work represents the first implementation of generative RBM inference on a neuromorphic VLSI substrate.

Artificial brains. A million spiking-neuron integrated circuit with a scalable communication network and interface.

Inspired by the brain's structure, we have developed an efficient, scalable, and flexible non-von Neumann architecture that leverages contemporary silicon technology. To demonstrate, we built a 5.4-billion-transistor chip with 4096 neurosynaptic cores interconnected via an intrachip network that integrates 1 million programmable spiking neurons and 256 million configurable synapses. Chips can be tiled in two dimensions via an interchip communication interface, seamlessly scaling the architecture to a cortexlike sheet of arbitrary size. The architecture is well suited to many applications that use complex neural networks in real time, for example, multiobject detection and classification. With 400-pixel-by-240-pixel video input at 30 frames per second, the chip consumes 63 milliwatts.

T-cell triggering thresholds are modulated by the number of antigen within individual T-cell receptor clusters.

T cells react to extremely small numbers of activating agonist peptides. Spatial organization of T-cell receptors (TCR) and their peptide-major histocompatibility complex (pMHC) ligands into microclusters is correlated with T-cell activation. Here we have designed an experimental strategy that enables control over the number of agonist peptides per TCR cluster, without altering the total number engaged by the cell. Supported membranes, partitioned with grids of barriers to lateral mobility, provide an effective way of limiting the total number of pMHC ligands that may be assembled within a single TCR cluster. Observations directly reveal that restriction of pMHC content within individual TCR clusters can decrease T-cell sensitivity for triggering initial calcium flux at fixed total pMHC density. Further analysis suggests that triggering thresholds are determined by the number of activating ligands available to individual TCR clusters, not by the total number encountered by the cell. Results from a series of experiments in which the overall agonist density and the maximum number of agonist per TCR cluster are independently varied in primary T cells indicate that the most probable minimal triggering unit for calcium signaling is at least four pMHC in a single cluster for this system. This threshold is unchanged by inclusion of coagonist pMHC, but costimulation of CD28 by CD80 can modulate the threshold lower.

Electrical manipulation of supported lipid membranes by embedded electrodes.

Alkanethiol modified gold electrodes patterned over a silica surface provided a dual hydrophobic/hydrophilic surface suitable for phospholipid monolayer and bilayer formation over the alkylated gold and glass surfaces, respectively. The phospholipid monolayer and bilayer were connected, allowing free diffusion of lipids within both leaflets of the glass-supported bilayer over the alkanethiol/gold-to-glass interface. Application of large alternating current fields to these electrodes irreversibly switched the gold electrodes to diffusion barriers. Enclosure of the electrode devices within protein barriers revealed a resting state surface potential driven reorganization of the charged fluorescent probes. Application of lower magnitude direct current fields resulted in electrophoretic redistribution of the membrane probes and electro-osmotic reorganization of membrane associated proteins.

Hybrid protein-lipid patterns from aluminum templates.

An aqueous aluminum liftoff process suitable for fabrication of hybrid patterns of protein and supported lipid membrane on silica surfaces is described. Patterned aluminum thin films, which can be produced by conventional optical or electron beam lithography, are employed as sacrificial protecting layers to define the geometry of the protein-lipid patterns. The aluminum is lifted off in a mildly basic aqueous solution, which preserves the integrity of bound protein layers. The newly exposed substrate can then be filled with supported membrane by exposure to an aqueous vesicle suspension. The final substrate consists of patterned protein and lipid membranes with spatial resolution determined by aluminum patterns, down to 200 nm line widths in this case. Inorganic surfaces were characterized by atomic force microscopy and X-ray photoelectron spectroscopy while supported bilayers and protein patterns were characterized by epifluorescence microscopy.

Scanning probe lithography on fluid lipid membranes.

Scanning probe lithography (SPL) is applied to pattern fluid lipid membranes on a solid borosilicate substrate. Grids of metal lines, prepatterned onto the substrate by electron beam lithography, serve to partition the supported membrane into an array of isolated fluid pixels. By toggling the pH of the surrounding solution, the effect of the probe tip on the membrane can be regulated. Alkaline conditions favor membrane removal, while neutral pH favors membrane deposition. Arbitrary membrane patterns with spatial dimensions limited by the underlying grid size can be constructed by sequential SPL membrane removal followed by refill with a different membrane type. In the present study, bilayers of unique composition fill 1 x 1 mum corrals and were positioned 100 nm apart.