An official website of the United States government
Spiking neural networks (SNNs) have received increasing attention due to their high biological plausibility and energy efficiency. The binary spike-based information propagation enables efficient sparse computation in event-based and static computer vision applications. However, the weight precision and especially the membrane potential precision remain as high-precision values (e.g., 32 bits) in state-of-the-art SNN algorithms. Each neuron in an SNN stores the membrane potential over time and typically updates its value in every time step. Such frequent read/write operations of high-precision membrane potential incur storage and memory access overhead in SNNs, which undermines the SNNs' compatibility with resource-constrained hardware. To resolve this inefficiency, prior works have explored the time step reduction and low-precision representation of membrane potential at a limited scale and reported significant accuracy drops. Furthermore, while recent advances in on-device AI present pruning and quantization optimization with different architectures and datasets, simultaneous pruning with quantization is highly under-explored in SNNs. In this work, we present SpQuant-SNN, a fully-quantized spiking neural network with ultra-low precision weights, membrane potential, and high spatial-channel sparsity, enabling the end-to-end low precision with significantly reduced operations on SNN. First, we propose an integer-only quantization scheme for the membrane potential with a stacked surrogate gradient function, a simple-yet-effective method that enables the smooth learning process of quantized SNN training. Second, we implement spatial-channel pruning with membrane potential prior, toward reducing the layer-wise computational complexity, and floating-point operations (FLOPs) in SNNs. Finally, to further improve the accuracy of low-precision and sparse SNN, we propose a self-adaptive learnable potential threshold for SNN training. Equipped with high biological adaptiveness, minimal computations, and memory utilization, SpQuant-SNN achieves state-of-the-art performance across multiple SNN models for both event-based and static image datasets, including both image classification and object detection tasks. The proposed SpQuant-SNN achieved up to 13× memory reduction and >4.7× FLOPs reduction with ~1.8% accuracy degradation for both classification and object detection tasks, compared to the SOTA baseline.
more »
« less
Ultra-thin and ultra-porous nanofiber networks as a basement-membrane mimic
A high porosity (88%) and ultrathin (<3 μm) fibrous basement membrane mimic using (A) suspended nanofiber networks for a (B) brain endothelial–pericyte co-culture model. (C) Our approach achieved low cell membrane and nuclei separations.
more »
« less
- PAR ID:
- 10469980
- Publisher / Repository:
- Lab on a Chip
- Date Published:
- Journal Name:
- Lab on a Chip
- Volume:
- 23
- Issue:
- 20
- ISSN:
- 1473-0197
- Page Range / eLocation ID:
- 4565 to 4578
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Thinning silicon wafers via wet etching is a common practice in the microelectromechanical system (MEMS) industry to produce membranes and other structures Wang (Nano Lett 13(9): 4393–4398, 2013). Controlling the thickness of a membrane is a critical aspect to optimize the functionality of these devices. Our research specifically focuses on the production of bio-membranes for lung-on-a-chip (LoaC) applications. In our fabrication, it is crucial for us to determine the membranes’ thickness in a non-invasive way. This study aims to address this issue by attempting to develop a tool that uses the optical properties of light transmission through silicon to find a correlation with thickness. To find this correlation, we conducted a small experimental study where we fabricated ultra-thin membranes and captured images of the light transmission through these samples. This paper will report the correlation found between calculated average intensities of these images and measurements done using scanning electron microscopy (SEM). Graphical abstractmore » « less
-
Water electrolysis using renewable energy inputs is being actively pursued as a green route for hydrogen production. However, it is limited by the high energy consumption due to the sluggish anodic oxygen evolution reaction (OER) and safety issues associated with H2 and O2 mixing. Here, we replaced OER with an electrocatalytic oxidative dehydrogenation (EOD) of aldehydes for bipolar H2 production and achieved industrial-level current densities at cell voltages much lower than during water electrolysis. Experimental and computational studies suggest a reasonable barrier for C-H dissociation on Cu surfaces, mainly through a diol intermediate, with a potential-dependent competition with the solution-phase Cannizzaro reaction. The kinetics of EOD reaction was further enhanced by a porous CuAg catalyst prepared from a galvanic replacement method. Through Ag incorporation and its modification of the Cu surface, the geometric current density and electrocatalyst durability were significantly improved. Finally, we engineered a bipolar H2 production system in membrane-electrode assembly-based flow cells to facilitate mass transport, achieving a maximum current density of 248 and 390 mA cm−2 at cell voltages of 0.4 V and 0.6 V, respectively. The faradaic efficiency of H2 from both cathode and anode reactions both attained ~100%. Taking advantage of the bipolar H2 production without the issues associated with H2/O2 mixing, an inexpensive, easy-to-manufacture dialysis porous membrane was demonstrated to substitute the costly anion exchange membrane, achieving an energy-efficient and cost-effective process in a simple reactor for H2 production. The estimated H2 price of $2.51/kg from an initial technoeconomic assessment is competitive with US DoE’s “Green H2” targets.more » « less
-
Abstract Current potentiometric sensing methods are limited to detecting nitrate at parts-per-billion (sub-micromolar) concentrations, and there are no existing potentiometric chemical sensors with ultralow detection limits below the parts-per-trillion (picomolar) level. To address these challenges, we integrate interdigital graphene ion-sensitive field-effect transistors (ISFETs) with a nitrate ion-sensitive membrane (ISM). The work aims to maximize nitrate ion transport through the nitrate ISM, while achieving high device transconductance by evaluating graphene layer thickness, optimizing channel width-to-length ratio (RWL), and enlarging total sensing area. The captured nitrate ions by the nitrate ISM induce surface potential changes that are transduced into electrical signals by graphene, manifested as the Dirac point shifts. The device exhibits Nernst response behavior under ultralow concentrations, achieving a sensitivity of 28 mV/decade and establishing a record low limit of detection of 0.041 ppt (4.8 × 10−13M). Additionally, the sensor showed a wide linear detection range from 0.1 ppt (1.2 × 10−12M) to 100 ppm (1.2 × 10−3M). Furthermore, successful detection of nitrate in tap and snow water was demonstrated with high accuracy, indicating promising applications to drinking water safety and environmental water quality control.more » « less
-
Abstract Yeasts are naturally diverse, genetically tractable, and easy to grow such that researchers can investigate any number of genotypes, environments, or interactions thereof. However, studies of yeast transcriptomes have been limited by the processing capabilities of traditional RNA sequencing techniques. Here we optimize a powerful, high‐throughput single‐cell RNA sequencing (scRNAseq) platform, SPLiT‐seq (Split Pool Ligation‐based Transcriptome sequencing), for yeasts and apply it to 43,388 cells of multiple species and ploidies. This platform utilizes a combinatorial barcoding strategy to enable massively parallel RNA sequencing of hundreds of yeast genotypes or growth conditions at once. This method can be applied to most species or strains of yeast for a fraction of the cost of traditional scRNAseq approaches. Thus, our technology permits researchers to leverage “the awesome power of yeast” by allowing us to survey the transcriptome of hundreds of strains and environments in a short period of time and with no specialized equipment. The key to this method is that sequential barcodes are probabilistically appended to cDNA copies of RNA while the molecules remain trapped inside of each cell. Thus, the transcriptome of each cell is labeled with a unique combination of barcodes. Since SPLiT‐seq uses the cell membrane as a container for this reaction, many cells can be processed together without the need to physically isolate them from one another in separate wells or droplets. Further, the first barcode in the sequence can be chosen intentionally to identify samples from different environments or genetic backgrounds, enabling multiplexing of hundreds of unique perturbations in a single experiment. In addition to greater multiplexing capabilities, our method also facilitates a deeper investigation of biological heterogeneity, given its single‐cell nature. For example, in the data presented here, we detect transcriptionally distinct cell states related to cell cycle, ploidy, metabolic strategies, and so forth, all within clonal yeast populations grown in the same environment. Hence, our technology has two obvious and impactful applications for yeast research: the first is the general study of transcriptional phenotypes across many strains and environments, and the second is investigating cell‐to‐cell heterogeneity across the entire transcriptome.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript
- Journal Article:
- https://doi.org/10.1039/d3lc00304c
