Search

Reverse engineering the brain by mimicking the structure and function of neuronal networks on a silicon integrated circuit was the original goal of neuromorphic engineering, but remains a distant prospect. The focus of neuromorphic engineering has thus been relaxed from rigorous brain mimicry to designs inspired by qualitative features of the brain, including event-driven signalling and in-memory information processing. Here we examine current approaches to neuromorphic engineering and provide a vision that returns neuromorphic electronics to its original goal of reverse engineering the brain. The essence of this vision is to ‘copy’ the functional synaptic connectivity map of a mammalian neuronal network using advanced neuroscience tools and then ‘paste’ this map onto a high-density three-dimensional network of solid-state memories. Our copy-and-paste approach could potentially lead to silicon integrated circuits that better approximate computing traits of the brain, including low power, facile learning, adaptation, and even autonomy and cognition. This Perspective explores the potential of an approach to neuromorphic electronics in which the functional synaptic connectivity map of a mammalian neuronal network is copied using a silicon neuro-electronic interface and then pasted onto a high-density three-dimensional network of solid-state memories.

Using ions in aqueous milieu for signal processing, like in biological circuits, may potentially lead to a bioinspired information processing platform. Studies, however, have focused on individual ionic diodes and transistors rather than circuits comprising many such devices. Here a 16 × 16 array of new ionic transistors is developed in an aqueous quinone solution. Each transistor features a concentric ring electrode pair with a disk electrode at the center. The electrochemistry of these electrodes in the solution provides the basis for the transistor operation. The ring pair electrochemically tunes the local electrolytic concentration to modulate the disk's Faradaic reaction rate. Thus, the disk current as a Faradaic reaction to the disk voltage is gated by the ring pair. The 16 × 16 array of these transistors performs analog multiply-accumulate (MAC) operations, a computing modality hotly pursued for low-power artificial neural networks. This exploits the transistor's operating regime where the disk current is a multiplication of the disk voltage and a weight parameter tuned by the ring pair gating. Such disk currents from multiple transistors are summated in a global reference electrode to complete a MAC task. This ionic circuit demonstrating analog computing is a step toward sophisticated aqueous ionics.

pH controls a large repertoire of chemical and biochemical processes in water. Densely arrayed pH microenvironments would parallelize these processes, enabling their high-throughput studies and applications. However, pH localization, let alone its arrayed realization, remains challenging because of fast diffusion of protons in water. Here, we demonstrate arrayed localizations of picoliter-scale aqueous acids, using a 256-electrochemical cell array defined on and operated by a complementary metal oxide semiconductor (CMOS)–integrated circuit. Each cell, comprising a concentric pair of cathode and anode with their current injections controlled with a sub-nanoampere resolution by the CMOS electronics, creates a local pH environment, or a pH “voxel,” via confined electrochemistry. The system also monitors the spatiotemporal pH profile across the array in real time for precision pH control. We highlight the utility of this CMOS pH localizer-imager for high-throughput tasks by parallelizing pH-gated molecular state encoding and pH-regulated enzymatic DNA elongation at any selected set of cells.

Electrode-based impedance and electrochemical measurements can provide cell-biology information that is difficult to obtain using optical-microscopy techniques. Such electrical methods are non-invasive, label-free, and continuous, eliminating the need for fluorescence reporters and overcoming optical imaging's throughput/temporal resolution limitations. Nonetheless, electrode-based techniques have not been heavily employed because devices typically contain few electrodes per well, resulting in noisy aggregate readouts. Complementary metal-oxide-semiconductor (CMOS) microelectrode arrays (MEAs) have sometimes been used for electrophysiological measurements with thousands of electrodes per well at sub-cellular pitches, but only basic impedance mappings of cell attachment have been performed outside of electrophysiology. Here, we report on new field-based impedance mapping and electrochemical mapping/patterning techniques to expand CMOS-MEA cell-biology applications. The methods enable accurate measurement of cell attachment, growth/wound healing, cell-cell adhesion, metabolic state, and redox properties with single-cell spatial resolution (20 μm electrode pitch). These measurements allow the quantification of adhesion and metabolic differences of cells expressing oncogenes

Portable NMR combining a permanent magnet and a complementary metal-oxide-semiconductor (CMOS) integrated circuit has recently emerged to offer the long desired online, on-demand, or in situ NMR analysis of small molecules for chemistry and biology. Here we take this cutting-edge technology to the next level by introducing parallelism to a state-of-the-art portable NMR platform to accelerate its experimental throughput, where NMR is notorious for inherently low throughput. With multiple (N) samples inside a single magnet, we perform simultaneous NMR analyses using a single silicon electronic chip, going beyond the traditional single-sample-per-magnet paradigm. We execute the parallel analyses via either time-interleaving or magnetic resonance imaging (MRI). In the time-interleaving method, the N samples occupy N separate NMR coils: we connect these N NMR coils to the single silicon chip one after another and repeat these sequential NMR scans. This time-interleaving is an effective parallelization, given a long recovery time of a single NMR scan. To demonstrate this time-interleaved parallelism, we use N = 2 for high-resolution multidimensional spectroscopy such as J-coupling resolved free induction decay spectroscopy and correlation spectroscopy (COSY) with the field homogeneity carefully optimized (

Complementary metal-oxide-semiconductor (CMOS) image sensors are a visual outpost of many machines that interact with the world. While they presently separate image capture in front-end silicon photodiode arrays from image processing in digital back-ends, efforts to process images within the photodiode array itself are rapidly emerging, in hopes of minimizing the data transfer between sensing and computing, and the associated overhead in energy and bandwidth. Electrical modulation, or programming, of photocurrents is requisite for such in-sensor computing, which was indeed demonstrated with electrostatically doped, but non-silicon, photodiodes. CMOS image sensors are currently incapable of in-sensor computing, as their chemically doped photodiodes cannot produce electrically tunable photocurrents. Here we report in-sensor computing with an array of electrostatically doped silicon p-i-n photodiodes, which is amenable to seamless integration with the rest of the CMOS image sensor electronics. This silicon-based approach could more rapidly bring in-sensor computing to the real world due to its compatibility with the mainstream CMOS electronics industry. Our wafer-scale production of thousands of silicon photodiodes using standard fabrication emphasizes this compatibility. We then demonstrate in-sensor processing of optical images using a variety of convolutional filters electrically programmed into a 3 × 3 network of these photodiodes.

Atomically thin two-dimensional (2D) materials-such as transition metal dichalcogenide (TMD) monolayers and hexagonal boron nitride (hBN)-and their van der Waals layered preparations have been actively researched to build electronic devices such as field-effect transistors, junction diodes, tunneling devices, and, more recently, memristors. Two-dimensional material memristors built in lateral form, with horizontal placement of electrodes and the 2D material layers, have provided an intriguing window into the motions of ions along the atomically thin layers. On the other hand, 2D material memristors built in vertical form with top and bottom electrodes sandwiching 2D material layers may provide opportunities to explore the extreme of the memristive performance with the atomic-scale interelectrode distance. In particular, they may help push the switching voltages to a lower limit, which is an important pursuit in memristor research in general, given their roles in neuromorphic computing. In fact, recently Akinwande et al. performed a pioneering work to demonstrate a vertical memristor that sandwiches a single MoS2 monolayer between two inert Au electrodes, but it could neither attain switching voltages below 1 V nor control the switching polarity, obtaining both unipolar and bipolar switching devices. Here, we report a vertical memristor that sandwiches two MoS2 monolayers between an active Cu top electrode and an inert Au bottom electrode. Cu ions diffuse through the MoS2 double layers to form atomic-scale filaments. The atomic-scale thickness, combined with the electrochemical metallization, lowers switching voltages down to 0.1-0.2 V, on par with the state of the art. Furthermore, our memristor achieves consistent bipolar and analogue switching, and thus exhibits the synapse-like learning behavior such as the spike-timing dependent plasticity (STDP), the very first STDP demonstration among all 2D-material-based vertical memristors. The demonstrated STDP with low switching voltages is promising not only for low-power neuromorphic computing, but also from the point of view that the voltage range approaches the biological action potentials, opening up a possibility for direct interfacing with mammalian neuronal networks.

Implementations of artificial neural networks that borrow analogue techniques could potentially offer low-power alternatives to fully digital approaches

The synaptic connections between neurons are traditionally determined by correlating the action potentials (APs) of a pre-synaptic neuron and small-amplitude subthreshold potentials of a post-synaptic neuron using invasive intracellular techniques, such as patch clamping. Extracellular recording by a microelectrode array can non-invasively monitor network activities of a large number of neurons, but its reduced sensitivity usually prevents direct measurements of synaptic signals. Here, we demonstrate that a newly developed complementary metal-oxide-semiconductor (CMOS) nanoelectrode array (CNEA) is capable of extracellularly determining direct synaptic connections in dense, multi-layer cultures of dissociated rat neurons. We spatiotemporally correlate action potential signals of hundreds of active neurons, detect small (∼1 pA after averaging) extracellular synaptic signals at the region where pre-synaptic axons and post-synaptic dendrites/somas overlap, and use those signals to map synaptic connections. We use controlled stimulation to assess stimulation-dependent synaptic strengths and to titrate a synaptic blocker (CNQX: IC50 ∼ 1 μM). The new capabilities demonstrated here significantly enhance the utilities of CNEAs in connectome mapping and drug screening applications.

CMOS microelectrode arrays (MEAs) can record electrophysiological activities of a large number of neurons in parallel but only extracellularly with a low signal-to-noise ratio. Patch-clamp electrodes can perform intracellular recording with a high signal-to-noise ratio but only from a few neurons in parallel. Recently, we have developed and reported a neuroelectronic interface that combines the parallelism of the CMOS MEA and the intracellular sensitivity of the patch clamp. Here, we report the design and characterization of the CMOS integrated circuit (IC), a critical component of the neuroelectronic interface. Fabricated in 0.18- $\mu \text{m}$ technology, the IC features an array of 4096 platinum black (PtB) nanoelectrodes spaced at a 20- $\mu \text{m}$ pitch on its surface and contains 4096 active pixel circuits. Each active pixel circuit, consisting of a new switched-capacitor current injector—-capable of injecting from ±15 pA to ±0.7 $\mu \text{A}$ with a 5-pA resolution—-and an operational amplifier, is highly configurable. When configured into the current-clamp mode, the pixel intracellularly records membrane potentials, including subthreshold activities with ~23- $\mu \text{V}_{\mathrm {rms}}$ input-referred noise while injecting a current for simultaneous stimulation. When configured into the voltage-clamp mode, the pixel becomes a switched-capacitor transimpedance amplifier with ~1-pArms input-referred noise and intracellularly records ion channel currents while applying a voltage for simultaneous stimulation. Such voltage-/current-clamp intracellular recording/stimulation is a feat only previously possible with the patch-clamp method. At the same time, as an array, the IC overcomes the lack of parallelism of the patch-clamp method, measuring thousands of mammalian neurons in parallel, with full-frame intracellular recording/stimulation at 9.4 kHz.

Developing a new tool capable of high-precision electrophysiological recording of a large network of electrogenic cells has long been an outstanding challenge in neurobiology and cardiology. Here, we combine nanoscale intracellular electrodes with complementary metal-oxide-semiconductor (CMOS) integrated circuits to realize a high-fidelity all-electrical electrophysiological imager for parallel intracellular recording at the network level. Our CMOS nanoelectrode array has 1,024 recording/stimulation 'pixels' equipped with vertical nanoelectrodes, and can simultaneously record intracellular membrane potentials from hundreds of connected in vitro neonatal rat ventricular cardiomyocytes. We demonstrate that this network-level intracellular recording capability can be used to examine the effect of pharmaceuticals on the delicate dynamics of a cardiomyocyte network, thus opening up new opportunities in tissue-based pharmacological screening for cardiac and neuronal diseases as well as fundamental studies of electrogenic cells and their networks.

Atomically thin two-dimensional (2D) materials-such as transition metal dichalcogenide (TMD) monolayers and hexagonal boron nitride (hBN)-and their van der Waals layered preparations have been actively researched to build electronic devices such as field-effect transistors, junction diodes, tunneling devices, and, more recently, memristors. Two-dimensional material memristors built in lateral form, with horizontal placement of electrodes and the 2D material layers, have provided an intriguing window into the motions of ions along the atomically thin layers. On the other hand, 2D material memristors built in vertical form with top and bottom electrodes sandwiching 2D material layers may provide opportunities to explore the extreme of the memristive performance with the atomic-scale interelectrode distance. In particular, they may help push the switching voltages to a lower limit, which is an important pursuit in memristor research in general, given their roles in neuromorphic computing. In fact, recently Akinwande et al. performed a pioneering work to demonstrate a vertical memristor that sandwiches a single MoS

The large-area synthesis of high-quality MoS

Ion-sensitive field-effect transistors (ISFETs) have been used to detect a variety of biomolecules whose charges alter the current or threshold voltage of the transistor. ISFETs can be built into large-scale arrays by CMOS compatible technology, offering the promise of highly parallel, all-electrical biomolecule sensing in a true chip-scale construct. Although CMOS-ISFET-based biomolecule detection has been amply demonstrated, a comprehensive optimization study of its sensitivity based on the device design, which would aid the development of large-scale arrays, remains lacking. Here, we present a systematic optimization strategy for CMOS-ISFET-based biomolecule sensing, using real-time DNA hybridization detection as an example. We analytically show that biasing the ISFETs close to the threshold is optimal, whereas minimizing the channel-to-sensing area ratio is beneficial but with limited sensitivity enhancement due to the double-layer capacitance of the sensing area. We then experimentally confirm our strategy with 26 ISFETs of varying sizes and biases from two CMOS chips, which detect the same 200-nM target DNA with different hybridization signals. The measured data correlate well to the presented theory. Our results are generally applicable to detecting other types of biomolecules, and may help in developing large-scale arrays of electrical biomolecular sensors.

2D semiconductors, especially transition metal dichalcogenide (TMD) monolayers, are extensively studied for electronic and optoelectronic applications. Beyond intensive studies on single transistors and photodetectors, the recent advent of large-area synthesis of these atomically thin layers has paved the way for 2D integrated circuits, such as digital logic circuits and image sensors, achieving an integration level of ≈100 devices thus far. Here, a decisive advance in 2D integrated circuits is reported, where the device integration scale is increased by tenfold and the functional complexity of 2D electronics is propelled to an unprecedented level. Concretely, an analog optoelectronic processor inspired by biological vision is developed, where 32 × 32 = 1024 MoS2 photosensitive field-effect transistors manifesting persistent photoconductivity (PPC) effects are arranged in a crossbar array. This optoelectronic processor with PPC memory mimics two core functions of human vision: it captures and stores an optical image into electrical data, like the eye and optic nerve chain, and then recognizes this electrical form of the captured image, like the brain, by executing analog in-memory neural net computing. In the highlight demonstration, the MoS2 FET crossbar array optically images 1000 handwritten digits and electrically recognizes these imaged data with 94% accuracy.