Probabilistic Neural Computing with Stochastic Devices.

The brain has effectively proven a powerful inspiration for the development of computing architectures in which processing is tightly integrated with memory, communication is event-driven, and analog computation can be performed at scale. These neuromorphic systems increasingly show an ability to improve the efficiency and speed of scientific computing and artificial intelligence applications. Herein, it is proposed that the brain's ubiquitous stochasticity represents an additional source of inspiration for expanding the reach of neuromorphic computing to probabilistic applications. To date, many efforts exploring probabilistic computing have focused primarily on one scale of the microelectronics stack, such as implementing probabilistic algorithms on deterministic hardware or developing probabilistic devices and circuits with the expectation that they will be leveraged by eventual probabilistic architectures. A co-design vision is described by which large numbers of devices, such as magnetic tunnel junctions and tunnel diodes, can be operated in a stochastic regime and incorporated into a scalable neuromorphic architecture that can impact a number of probabilistic computing applications, such as Monte Carlo simulations and Bayesian neural networks. Finally, a framework is presented to categorize increasingly advanced hardware-based probabilistic computing technologies.

Magnetic Source Imaging Using a Pulsed Optically Pumped Magnetometer Array.

We have developed a pulsed optically pumped magnetometer (OPM) array for detecting magnetic field maps originated from an arbitrary current distribution. The presented magnetic source imaging (MSI) system features 24 OPM channels, has a data rate of 500 S/s, a sensitivity of 0.8 p T / H z , and a dynamic range of 72 dB. We have employed our pulsed- OPM MSI system for measuring the magnetic field map of a test coil structure. The coils are moved across the array in an indexed fashion to measure the magnetic field over an area larger than the array. The captured magnetic field maps show excellent agreement with the simulation results. Assuming a 2D current distribution, we have solved the inverse problem, using the measured magnetic field maps, and the reconstructed current distribution image is compared to that of the simulation.

Parallel programming of an ionic floating-gate memory array for scalable neuromorphic computing.

Neuromorphic computers could overcome efficiency bottlenecks inherent to conventional computing through parallel programming and readout of artificial neural network weights in a crossbar memory array. However, selective and linear weight updates and <10-nanoampere read currents are required for learning that surpasses conventional computing efficiency. We introduce an ionic floating-gate memory array based on a polymer redox transistor connected to a conductive-bridge memory (CBM). Selective and linear programming of a redox transistor array is executed in parallel by overcoming the bridging threshold voltage of the CBMs. Synaptic weight readout with currents <10 nanoamperes is achieved by diluting the conductive polymer with an insulator to decrease the conductance. The redox transistors endure >1 billion write-read operations and support >1-megahertz write-read frequencies.

Computing with Spikes: The Advantage of Fine-Grained Timing.

Neural-inspired spike-based computing machines often claim to achieve considerable advantages in terms of energy and time efficiency by using spikes for computation and communication. However, fundamental questions about spike-based computation remain unanswered. For instance, how much advantage do spike-based approaches have over conventional methods, and under what circumstances does spike-based computing provide a comparative advantage? Simply implementing existing algorithms using spikes as the medium of computation and communication is not guaranteed to yield an advantage. Here, we demonstrate that spike-based communication and computation within algorithms can increase throughput, and they can decrease energy cost in some cases. We present several spiking algorithms, including sorting a set of numbers in ascending/descending order, as well as finding the maximum or minimum or median of a set of numbers. We also provide an example application: a spiking median-filtering approach for image processing providing a low-energy, parallel implementation. The algorithms and analyses presented here demonstrate that spiking algorithms can provide performance advantages and offer efficient computation of fundamental operations useful in more complex algorithms.

Rapid Nucleic Acid Extraction and Purification Using a Miniature Ultrasonic Technique.

Miniature ultrasonic lysis for biological sample preparation is a promising technique for efficient and rapid extraction of nucleic acids and proteins from a wide variety of biological sources. Acoustic methods achieve rapid, unbiased, and efficacious disruption of cellular membranes while avoiding the use of harsh chemicals and enzymes, which interfere with detection assays. In this work, a miniature acoustic nucleic acid extraction system is presented. Using a miniature bulk acoustic wave (BAW) transducer array based on 36° Y-cut lithium niobate, acoustic waves were coupled into disposable laminate-based microfluidic cartridges. To verify the lysing effectiveness, the amount of liberated ATP and the cell viability were measured and compared to untreated samples. The relationship between input power, energy dose, flow-rate, and lysing efficiency were determined. DNA was purified on-chip using three approaches implemented in the cartridges: a silica-based sol-gel silica-bead filled microchannel, nucleic acid binding magnetic beads, and Nafion-coated electrodes. Using E. coli, the lysing dose defined as ATP released per joule was 2.2× greater, releasing 6.1× more ATP for the miniature BAW array compared to a bench-top acoustic lysis system. An electric field-based nucleic acid purification approach using Nafion films yielded an extraction efficiency of 69.2% in 10 min for 50 µL samples.

A Combinatorial Model for Dentate Gyrus Sparse Coding.

The dentate gyrus forms a critical link between the entorhinal cortex and CA3 by providing a sparse version of the signal. Concurrent with this increase in sparsity, a widely accepted theory suggests the dentate gyrus performs pattern separation-similar inputs yield decorrelated outputs. Although an active region of study and theory, few logically rigorous arguments detail the dentate gyrus's (DG) coding. We suggest a theoretically tractable, combinatorial model for this action. The model provides formal methods for a highly redundant, arbitrarily sparse, and decorrelated output signal.To explore the value of this model framework, we assess how suitable it is for two notable aspects of DG coding: how it can handle the highly structured grid cell representation in the input entorhinal cortex region and the presence of adult neurogenesis, which has been proposed to produce a heterogeneous code in the DG. We find tailoring the model to grid cell input yields expansion parameters consistent with the literature. In addition, the heterogeneous coding reflects activity gradation observed experimentally. Finally, we connect this approach with more conventional binary threshold neural circuit models via a formal embedding.

Li-Ion Synaptic Transistor for Low Power Analog Computing.

Nonvolatile redox transistors (NVRTs) based upon Li-ion battery materials are demonstrated as memory elements for neuromorphic computer architectures with multi-level analog states, "write" linearity, low-voltage switching, and low power dissipation. Simulations of backpropagation using the device properties reach ideal classification accuracy. Physics-based simulations predict energy costs per "write" operation of <10 aJ when scaled to 200 nm × 200 nm.

Single objective light-sheet microscopy for high-speed whole-cell 3D super-resolution.

We have developed a method for performing light-sheet microscopy with a single high numerical aperture lens by integrating reflective side walls into a microfluidic chip. These 45° side walls generate light-sheet illumination by reflecting a vertical light-sheet into the focal plane of the objective. Light-sheet illumination of cells loaded in the channels increases image quality in diffraction limited imaging via reduction of out-of-focus background light. Single molecule super-resolution is also improved by the decreased background resulting in better localization precision and decreased photo-bleaching, leading to more accepted localizations overall and higher quality images. Moreover, 2D and 3D single molecule super-resolution data can be acquired faster by taking advantage of the increased illumination intensities as compared to wide field, in the focused light-sheet.

Induction frequency affects cortico-striatal synaptic plasticity with implications for frequency filtering.

Long-term synaptic depression (LTD) in cortico-striatal circuits is initiated by depolarization of striatal medium spiny neurons through a convergent cortical glutamatergic input. This produces retrograde endocannabinoid signaling to presynaptic cortical terminals and eventually results in long term (>30 min) decreases in glutamate release. These same circuits can also undergo short-term depression (STD) through a less well-defined process in which the magnitude of postsynaptic responses returns to baseline levels within 10 min. Additionally, the cortico-striatal circuit shows characteristics of a GABAA receptor-dependent low-pass filter, which results in significant attenuation of high frequency cortical inputs. The majority of in vitro studies of LTD have used a 100-Hz induction paradigm and it is unclear whether other frequencies, which may also have physiological relevance, have equivalent ability to induce this form of plasticity. Here we have investigated the effectiveness of a range of induction paradigms in producing LTD in cortico-striatal circuits, and demonstrate that some lower frequency paradigms, with perhaps more physiological relevance, are more effective at inducing LTD. We also show that GABAA receptor-dependent frequency filtering in this circuit is altered following the induction of LTD and STD suggesting an important role for synaptic depression in signal processing in these circuits.

Energy Scaling Advantages of Resistive Memory Crossbar Based Computation and Its Application to Sparse Coding.

The exponential increase in data over the last decade presents a significant challenge to analytics efforts that seek to process and interpret such data for various applications. Neural-inspired computing approaches are being developed in order to leverage the computational properties of the analog, low-power data processing observed in biological systems. Analog resistive memory crossbars can perform a parallel read or a vector-matrix multiplication as well as a parallel write or a rank-1 update with high computational efficiency. For an N × N crossbar, these two kernels can be O(N) more energy efficient than a conventional digital memory-based architecture. If the read operation is noise limited, the energy to read a column can be independent of the crossbar size (O(1)). These two kernels form the basis of many neuromorphic algorithms such as image, text, and speech recognition. For instance, these kernels can be applied to a neural sparse coding algorithm to give an O(N) reduction in energy for the entire algorithm when run with finite precision. Sparse coding is a rich problem with a host of applications including computer vision, object tracking, and more generally unsupervised learning.

Isothermal switching and detailed filament evolution in memristive systems.

The steady-state solution of filamentary memristive switching may be derived directly from the heat equation, modelling vertical and radial heat flow. This solution is shown to provide a continuous and accurate description of the evolution of the filament radius, composition, heat flow, and temperature during switching, and is shown to apply to a large range of switching materials and experimental time-scales.

Combined chemical and topographical guidance cues for directing cytoarchitectural polarization in primary neurons.

Chemical and topographical cues can be used to guide dissociated neurons into user-defined network geometries on artificial substrates, yet control of neuron polarity (differentiation into axons and dendrites) remains an elusive goal. We developed a dual guidance cue strategy for directing morphological maturity in neurons in vitro using combined chemical and topographical guidance cues on glass substrates. The surface chemistry provides chemical attraction and repulsion for controlling neuron placement and outgrowth, while the topography provides additional surface area for neuron attachment. Poly-l-lysine (PLL) was adsorbed into etched trenches in glass substrates, and an acetone liftoff process was used to produce bifunctional surfaces with a hydrophobic hexamethyldisilazane (HMDS) background and trench patterns of PLL. We examined the cytoarchitectural polarization of dissociated hippocampal pyramidal neurons on guidance cues designed to promote rapid outgrowth of neurites onto continuous line features and delayed neurite outgrowth onto interrupted line features. An optimum distance of approximately 5 µm between the cell body attachment node and the first interrupted line guidance cue led to specific cytoarchitectural polarization of ≥60% of neurons by 3 days of culture in vitro.

Nuclear translocation kinetics of NF-kappaB in macrophages challenged with pathogens in a microfluidic platform.

We have developed a microfluidic platform for real-time imaging of host-pathogen interactions and cellular signaling events. Host cells are immobilized in a controlled environment for optical interrogation of the kinetics and stochasticity of immune response to pathogenic challenges. Here, we have quantitatively measured activation of the toll-like receptor 4 (TLR4) pathway in RAW264.7 murine macrophage-like cells. This was achieved by measuring the cytoplasm-to-nucleus translocation kinetics of a green fluorescent protein fusion construct to the NF-kappaB transcription factor subunit RelA (GFP-RelA). Translocation kinetics in response to live bacteria and purified lipopolysaccharide (LPS) challenges were measured, and this work presents the first demonstration of live imaging of host cell infection on a microfluidic platform with quantitative analysis of an early (<0.5 h from infection) immune signaling event. Our data show that a 1,000x increase in the LPS dose led to a ~10x increase in a host cell activation metric we developed in order to describe NF-kappaB translocation kinetics. Using this metric, live bacteria challenges were assigned an equivalent LPS dose as a first step towards comparing NF-kappaB translocation kinetics between TLR4-only pathway signaling (activated by LPS) and multiple pathway signaling (activated by whole bacteria). The device also contains a unique architecture for capturing and fluidically isolating single host cells for the purpose of differentiating between primary and secondary immune signaling.

Impedimetric and optical interrogation of single cells in a microfluidic device for real-time viability and chemical response assessment.

We report here a non-invasive, reversible method for interrogating single cells in a microfluidic flow-through system. Impedance spectroscopy of cells held at a micron-sized pore under negative pressure is demonstrated and used to determine the presence and viability of the captured cell. The cell capture pore is optimized for electrical response and mechanical interfacing to a cell using a deposited layer of parylene. Changes in the mechanical interface between the cell and the chip due to chemical exposure or environmental changes can also be assessed. Here, we monitored the change in adhesion/spreading of RAW264.7 macrophages in response to the immune stimulant lipopolysaccharide (LPS). This method enables selective, reversible, and quantitative long-term impedance measurements on single cells. The fully sealed electrofluidic assembly is compatible with long-term cell culturing, and could be modified to incorporate single cell lysis and subsequent intracellular separation and analysis.

Effects of substrate geometry on growth cone behavior and axon branching.

At the leading edge of a growing axon, the growth cone determines the path the axon takes and also plays a role in the formation of branches, decisions that are regulated by a complex array of chemical signals. Here, we used microfabrication technology to determine whether differences in substrate geometry, independent of changes in substrate chemistry, can modulate growth cone motility and branching, by patterning a polylysine grid of narrow (2 or 5 microm wide) intersecting lines. The shape of the intersections varied from circular nodes 15 microm in diameter to simple crossed lines (nodeless intersections). Time-lapse recordings of cultured hippocampal neurons showed that simple variations in substrate geometry changed growth cone shape, and altered the rate of growth and the probability of branching. When crossing onto a node intersection the growth cone paused, often for hours, and microtubules appeared to defasciculate. Once beyond the node, filopodia and lamellipodia persisted at that site, sometimes forming a collateral branch. At nodeless intersections, the growth cone passed through with minimal hesitation, often becoming divided into separate areas of motility that led to the growth of separate branches. When several lines intersected at a common point, growth cones sometimes split into several subdivisions, resulting in the emergence of as many as five branches. Such experiments revealed an intrinsic preference for branches to form at angles less than 90 degrees . These data show that simple changes in the geometry of a chemically homogeneous substrate are detected by the growth cone and can regulate axonal growth and the formation of branches.

Extracellular recordings from patterned neuronal networks using planar microelectrode arrays.

Neuronal cell networks have been reconstructed on planar microelectrode arrays (MEAs) from dissociated hippocampal pyramidal neurons. Microcontact printing (microCP) and a photoresist-liftoff method were used to selectively localize poly-L-lysine (PLL) on the surface of MEAs. Haptotaxis led to the organization of the neurons into networks localized adjacent to microelectrodes. Various grids of PLL with 2-25-microm-wide lines spaced by 50-200 microm with 15-25-microm nodes at intersection points were used to guide cell body attachment and neurite outgrowth. Bursting activity with spike amplitude attenuation was observed, and multichannel recordings detected instances of coincident firing activity. Finally, we present here an extracellular recording from a approximately 2 microm bundle of guided neurites.

Patterning axonal guidance molecules using a novel strategy for microcontact printing.

We present here a two-step strategy for micropatterning proteins on a substrate to control neurite growth in culture. First, conventional microcontact printing is used to prepare a micropattern of protein A, which binds the Fc fragment of immunoglobulins. Then, a chimeric protein, consisting of the extracellular domain of a guidance protein recombinantly linked to the Fc fragment of IgG (prepared using conventional molecular techniques), is applied from solution. The chimeric protein binds to the patterned protein A, taking on its geometric pattern. Using this method, we have micropatterned the extracellular domain of the cell adhesion molecule, L1 (as an L1-Fc chimera) and demonstrated that it retains its ability to selectively guide axonal growth. L1-Fc micropatterned on a background of poly-L-lysine resulted in selective growth of the axons on the micropattern, whereas the somata and dendrites were unresponsive. Substrates bearing simultaneous micropatterns of L1-Fc and poly-L-lysine on a background of untreated glass were also created. Using this approach, cell body position was controlled by manipulating the dimensions of the poly-L-lysine pattern, and the dendrites were constrained to the poly-L-lysine pattern, while the axons grew preferentially on L1-Fc. The two-step microcontact printing method allows preparation of substrates that contain guidance proteins in geometric patterns with resolution of approximately 1 microm. This method should be broadly applicable to many classes of proteins.