arXiv subjects

Aayush Ankit

Publications and source records attributed to Aayush Ankit.

17 recordsLinked to original sources

arXiv2026

Unveiling the Potential of Quantization with MXFP4: Strategies for Quantization Error Reduction

Large Language Models (LLMs) have intensified the need for low-precision formats that enable efficient, large-scale inference. The Open Compute Project (OCP) Microscaling (MX) standard is attractive due to its favorable hardware efficiency, but its 4-bit variant (MXFP4) lags behind NVIDIA's NVFP4 in accuracy, limiting adoption. We introduce two software-only techniques, Overflow-Aware Scaling (OAS) and Macro Block Scaling (MBS), that improve MXFP4 quantization fidelity without requiring hardware changes. OAS reduces overall errors by increasing effective dynamic range under power-of-two block scaling, while MBS allocates higher-precision scaling at a coarser granularity to better preserve outliers. Across multiple LLMs and standard downstream benchmarks, OAS and MBS reduce the end-to-end accuracy gap between MXFP4 and NVFP4 from about 10% to below 1% on average, while incurring modest GEMM overhead (6.2% on average). These results re-establish MXFP4 as a practical alternative to NVFP4, enabling near-NVFP4 accuracy while retaining MX's hardware-efficiency advantages (e.g., 12% relative area savings in tensor cores).

cs.AR↗

arXiv2025

COMET: A Framework for Modeling Compound Operation Dataflows with Explicit Collectives

Modern machine learning accelerators are designed to efficiently execute deep neural networks (DNNs) by optimizing data movement, memory hierarchy, and compute throughput. However, emerging DNN models such as large language models, state space models increasingly rely on compound operations-structured compositions of multiple basic operations-which introduce new challenges for dataflow optimization and minimizing off-chip memory traffic. Moreover, as model size continues to grow, deployment across spatially distributed compute clusters becomes essential, requiring frequent and complex collective communication. Existing dataflow optimization frameworks and performance models either focus on single operations or lack explicit modeling of collective communication cost, limiting their applicability to modern workloads. To address these limitations, we propose, a framework for modeling and optimizing dataflow for compound operations on machine learning accelerators. COMET introduces a novel representation that explicitly models collective communication across spatial clusters, along with latency and energy cost models that account for both GEMM and non-GEMM operation level dependencies within compound operations. We demonstrate COMET's capabilities to analyze and optimize dataflows for compound operations such as GEMM--Softmax, GEMM--LayerNorm, and self-attention, across both edge and cloud accelerator configurations. Our collective-aware modeling enables exploration of a broader mapping space, leading to improved performance and energy efficiency. Specifically, our optimized dataflows achieve up to 1.42$\times$ speedup for GEMM-Softmax, 3.46$\times$ for GEMM-LayerNorm and 1.82$\times$ for self-attention compared to unfused baselines.

cs.AR↗

arXiv2021

NAX: Co-Designing Neural Network and Hardware Architecture for Memristive Xbar based Computing Systems

In-Memory Computing (IMC) hardware using Memristive Crossbar Arrays (MCAs) are gaining popularity to accelerate Deep Neural Networks (DNNs) since it alleviates the "memory wall" problem associated with von-Neumann architecture. The hardware efficiency (energy, latency and area) as well as application accuracy (considering device and circuit non-idealities) of DNNs mapped to such hardware are co-dependent on network parameters, such as kernel size, depth etc. and hardware architecture parameters such as crossbar size. However, co-optimization of both network and hardware parameters presents a challenging search space comprising of different kernel sizes mapped to varying crossbar sizes. To that effect, we propose NAX -- an efficient neural architecture search engine that co-designs neural network and IMC based hardware architecture. NAX explores the aforementioned search space to determine kernel and corresponding crossbar sizes for each DNN layer to achieve optimal tradeoffs between hardware efficiency and application accuracy. Our results from NAX show that the networks have heterogeneous crossbar sizes across different network layers, and achieves optimal hardware efficiency and accuracy considering the non-idealities in crossbars. On CIFAR-10 and Tiny ImageNet, our models achieve 0.8%, 0.2% higher accuracy, and 17%, 4% lower EDAP (energy-delay-area product) compared to a baseline ResNet-20 and ResNet-18 models, respectively.

cs.ET↗

arXiv2021

SPACE: Structured Compression and Sharing of Representational Space for Continual Learning

Humans learn adaptively and efficiently throughout their lives. However, incrementally learning tasks causes artificial neural networks to overwrite relevant information learned about older tasks, resulting in 'Catastrophic Forgetting'. Efforts to overcome this phenomenon often utilize resources poorly, for instance, by growing the network architecture or needing to save parametric importance scores, or violate data privacy between tasks. To tackle this, we propose SPACE, an algorithm that enables a network to learn continually and efficiently by partitioning the learnt space into a Core space, that serves as the condensed knowledge base over previously learned tasks, and a Residual space, which is akin to a scratch space for learning the current task. After learning each task, the Residual is analyzed for redundancy, both within itself and with the learnt Core space. A minimal number of extra dimensions required to explain the current task are added to the Core space and the remaining Residual is freed up for learning the next task. We evaluate our algorithm on P-MNIST, CIFAR and a sequence of 8 different datasets, and achieve comparable accuracy to the state-of-the-art methods while overcoming catastrophic forgetting. Additionally, our algorithm is well suited for practical use. The partitioning algorithm analyzes all layers in one shot, ensuring scalability to deeper networks. Moreover, the analysis of dimensions translates to filter-level sparsity, and the structured nature of the resulting architecture gives us up to 5x improvement in energy efficiency during task inference over the current state-of-the-art.

cs.LG↗

arXiv2020

GENIEx: A Generalized Approach to Emulating Non-Ideality in Memristive Xbars using Neural Networks

The analog nature of computing in Memristive crossbars poses significant issues due to various non-idealities such as: parasitic resistances, non-linear I-V characteristics of the device etc. The non-idealities can have a detrimental impact on the functionality i.e. computational accuracy of crossbars. Past works have explored modeling the non-idealities using analytical techniques. However, several non-idealities have data dependent behavior. This can not be captured using analytical (non data-dependent) models thereby, limiting their suitability in predicting application accuracy. To address this, we propose a Generalized Approach to Emulating Non-Ideality in Memristive Crossbars using Neural Networks (GENIEx), which accurately captures the data-dependent nature of non-idealities. We perform extensive HSPICE simulations of crossbars with different voltage and conductance combinations. Following that, we train a neural network to learn the transfer characteristics of the non-ideal crossbar. Next, we build a functional simulator which includes key architectural facets such as \textit{tiling}, and \textit{bit-slicing} to analyze the impact of non-idealities on the classification accuracy of large-scale neural networks. We show that GENIEx achieves \textit{low} root mean square errors (RMSE) of $0.25$ and $0.7$ for low and high voltages, respectively, compared to HSPICE. Additionally, the GENIEx errors are $7\times$ and $12.8\times$ better than an analytical model which can only capture the linear non-idealities. Further, using the functional simulator and GENIEx, we demonstrate that an analytical model can overestimate the degradation in classification accuracy by $\ge 10\%$ on CIFAR-100 and $3.7\%$ on ImageNet datasets compared to GENIEx.

cs.ET↗

arXiv2019

PANTHER: A Programmable Architecture for Neural Network Training Harnessing Energy-efficient ReRAM

The wide adoption of deep neural networks has been accompanied by ever-increasing energy and performance demands due to the expensive nature of training them. Numerous special-purpose architectures have been proposed to accelerate training: both digital and hybrid digital-analog using resistive RAM (ReRAM) crossbars. ReRAM-based accelerators have demonstrated the effectiveness of ReRAM crossbars at performing matrix-vector multiplication operations that are prevalent in training. However, they still suffer from inefficiency due to the use of serial reads and writes for performing the weight gradient and update step. A few works have demonstrated the possibility of performing outer products in crossbars, which can be used to realize the weight gradient and update step without the use of serial reads and writes. However, these works have been limited to low precision operations which are not sufficient for typical training workloads. Moreover, they have been confined to a limited set of training algorithms for fully-connected layers only. To address these limitations, we propose a bit-slicing technique for enhancing the precision of ReRAM-based outer products, which is substantially different from bit-slicing for matrix-vector multiplication only. We incorporate this technique into a crossbar architecture with three variants catered to different training algorithms. To evaluate our design on different types of layers in neural networks (fully-connected, convolutional, etc.) and training algorithms, we develop PANTHER, an ISA-programmable training accelerator with compiler support. Our evaluation shows that PANTHER achieves up to $8.02\times$, $54.21\times$, and $103\times$ energy reductions as well as $7.16\times$, $4.02\times$, and $16\times$ execution time reductions compared to digital accelerators, ReRAM-based accelerators, and GPUs, respectively.

cs.DC↗

arXiv2019

Constructing Energy-efficient Mixed-precision Neural Networks through Principal Component Analysis for Edge Intelligence

The `Internet of Things' has brought increased demand for AI-based edge computing in applications ranging from healthcare monitoring systems to autonomous vehicles. Quantization is a powerful tool to address the growing computational cost of such applications, and yields significant compression over full-precision networks. However, quantization can result in substantial loss of performance for complex image classification tasks. To address this, we propose a Principal Component Analysis (PCA) driven methodology to identify the important layers of a binary network, and design mixed-precision networks. The proposed Hybrid-Net achieves a more than 10% improvement in classification accuracy over binary networks such as XNOR-Net for ResNet and VGG architectures on CIFAR-100 and ImageNet datasets while still achieving up to 94% of the energy-efficiency of XNOR-Nets. This work furthers the feasibility of using highly compressed neural networks for energy-efficient neural computing in edge devices.

cs.LG↗

arXiv2019

PABO: Pseudo Agent-Based Multi-Objective Bayesian Hyperparameter Optimization for Efficient Neural Accelerator Design

The ever increasing computational cost of Deep Neural Networks (DNN) and the demand for energy efficient hardware for DNN acceleration has made accuracy and hardware cost co-optimization for DNNs tremendously important, especially for edge devices. Owing to the large parameter space and cost of evaluating each parameter in the search space, manually tuning of DNN hyperparameters is impractical. Automatic joint DNN and hardware hyperparameter optimization is indispensable for such problems. Bayesian optimization-based approaches have shown promising results for hyperparameter optimization of DNNs. However, most of these techniques have been developed without considering the underlying hardware, thereby leading to inefficient designs. Further, the few works that perform joint optimization are not generalizable and mainly focus on CMOS-based architectures. In this work, we present a novel pseudo agent-based multi-objective hyperparameter optimization (PABO) for maximizing the DNN performance while obtaining low hardware cost. Compared to the existing methods, our work poses a theoretically different approach for joint optimization of accuracy and hardware cost and focuses on memristive crossbar-based accelerators. PABO uses a supervisor agent to establish connections between the posterior Gaussian distribution models of network accuracy and hardware cost requirements. The agent reduces the mathematical complexity of the co-optimization problem by removing unnecessary computations and updates of acquisition functions, thereby achieving significant speed-ups for the optimization procedure. PABO outputs a Pareto frontier that underscores the trade-offs between designing high-accuracy and hardware efficiency. Our results demonstrate a superior performance compared to the state-of-the-art methods both in terms of accuracy and computational speed (~100x speed up).

cs.LG↗

arXiv2019

Incremental Learning in Deep Convolutional Neural Networks Using Partial Network Sharing

Deep convolutional neural network (DCNN) based supervised learning is a widely practiced approach for large-scale image classification. However, retraining these large networks to accommodate new, previously unseen data demands high computational time and energy requirements. Also, previously seen training samples may not be available at the time of retraining. We propose an efficient training methodology and incrementally growing DCNN to learn new tasks while sharing part of the base network. Our proposed methodology is inspired by transfer learning techniques, although it does not forget previously learned tasks. An updated network for learning new set of classes is formed using previously learned convolutional layers (shared from initial part of base network) with addition of few newly added convolutional kernels included in the later layers of the network. We employed a `clone-and-branch' technique which allows the network to learn new tasks one after another without any performance loss in old tasks. We evaluated the proposed scheme on several recognition applications. The classification accuracy achieved by our approach is comparable to the regular incremental learning approach (where networks are updated with new training samples only, without any network sharing), while achieving energy efficiency, reduction in storage requirements, memory access and training time.

cs.CV↗

arXiv2019

Efficient Hybrid Network Architectures for Extremely Quantized Neural Networks Enabling Intelligence at the Edge

The recent advent of `Internet of Things' (IOT) has increased the demand for enabling AI-based edge computing. This has necessitated the search for efficient implementations of neural networks in terms of both computations and storage. Although extreme quantization has proven to be a powerful tool to achieve significant compression over full-precision networks, it can result in significant degradation in performance. In this work, we propose extremely quantized hybrid network architectures with both binary and full-precision sections to emulate the classification performance of full-precision networks while ensuring significant energy efficiency and memory compression. We explore several hybrid network architectures and analyze the performance of the networks in terms of accuracy, energy efficiency and memory compression. We perform our analysis on ResNet and VGG network architectures. Among the proposed network architectures, we show that the hybrid networks with full-precision residual connections emerge as the optimum by attaining accuracies close to full-precision networks while achieving excellent memory compression, up to 21.8x in case of VGG-19. This work demonstrates an effective way of hybridizing networks which achieve performance close to full-precision networks while attaining significant compression, furthering the feasibility of using such networks for energy-efficient neural computing in IOT-based edge devices.

cs.LG↗

arXiv2019

PUMA: A Programmable Ultra-efficient Memristor-based Accelerator for Machine Learning Inference

Memristor crossbars are circuits capable of performing analog matrix-vector multiplications, overcoming the fundamental energy efficiency limitations of digital logic. They have been shown to be effective in special-purpose accelerators for a limited set of neural network applications. We present the Programmable Ultra-efficient Memristor-based Accelerator (PUMA) which enhances memristor crossbars with general purpose execution units to enable the acceleration of a wide variety of Machine Learning (ML) inference workloads. PUMA's microarchitecture techniques exposed through a specialized Instruction Set Architecture (ISA) retain the efficiency of in-memory computing and analog circuitry, without compromising programmability. We also present the PUMA compiler which translates high-level code to PUMA ISA. The compiler partitions the computational graph and optimizes instruction scheduling and register allocation to generate code for large and complex workloads to run on thousands of spatial cores. We have developed a detailed architecture simulator that incorporates the functionality, timing, and power models of PUMA's components to evaluate performance and energy consumption. A PUMA accelerator running at 1 GHz can reach area and power efficiency of $577~GOPS/s/mm^2$ and $837~GOPS/s/W$, respectively. Our evaluation of diverse ML applications from image recognition, machine translation, and language modelling (5M-800M synapses) shows that PUMA achieves up to $2,446\times$ energy and $66\times$ latency improvement for inference compared to state-of-the-art GPUs. Compared to an application-specific memristor-based accelerator, PUMA incurs small energy overheads at similar inference latency and added programmability.

cs.ET↗

arXiv2018

Xcel-RAM: Accelerating Binary Neural Networks in High-Throughput SRAM Compute Arrays

Deep neural networks are a biologically-inspired class of algorithms that have recently demonstrated state-of-the-art accuracies involving large-scale classification and recognition tasks. Indeed, a major landmark that enables efficient hardware accelerators for deep networks is the recent advances from the machine learning community that have demonstrated aggressively scaled deep binary networks with state-of-the-art accuracies. In this paper, we demonstrate how deep binary networks can be accelerated in modified von-Neumann machines by enabling binary convolutions within the SRAM array. In general, binary convolutions consist of bit-wise XNOR followed by a population-count (popcount). We present a charge sharing XNOR and popcount operation in 10 transistor SRAM cells. We have employed multiple circuit techniques including dual-read-worldines (Dual-RWL) along with a dual-stage ADC that overcomes the inaccuracies of a low precision ADC, to achieve a fairly accurate popcount. In addition, a key highlight of the present work is the fact that we propose sectioning of the SRAM array by adding switches onto the read-bitlines, thereby achieving improved parallelism. This is beneficial for deep networks, where the kernel size grows and requires to be stored in multiple sub-banks. As such, one needs to evaluate the partial popcount from multiple sub-banks and sum them up for achieving the final popcount. For n-sections per sub-array, we can perform n convolutions within one particular sub-bank, thereby improving overall system throughput as well as the energy efficiency. Our results at the array level show that the energy consumption and delay per-operation was 1.914pJ and 45ns, respectively. Moreover, an energy improvement of 2.5x, and a performance improvement of 4x was achieved by using the proposed sectioned-SRAM, compared to a non-sectioned SRAM design.

cs.ET↗

arXiv2018

SPARE: Spiking Networks Acceleration Using CMOS ROM-Embedded RAM as an In-Memory-Computation Primitive

Despite huge success of artificial intelligence, hardware systems running these algorithms consume orders of magnitude higher energy compared to the human brain, mainly due to heavy data movements between the memory unit and the computation cores. Spiking neural networks (SNNs) built using bio-plausible neuron and synaptic models have emerged as the power-efficient choice for designing cognitive applications. These algorithms involve several lookup-table (LUT) based function evaluations such as high-order polynomials and transcendental functions for solving complex neuro-synaptic models, that typically require additional storage. To that effect, we propose `SPARE' - an in-memory, distributed processing architecture built on ROM-embedded RAM technology, for accelerating SNNs. ROM-embedded RAMs allow storage of LUTs, embedded within a typical memory array, without additional area overhead. Our proposed architecture consists of a 2-D array of Processing Elements (PEs). Since most of the computations are done locally within each PE, unnecessary data transfers are restricted, thereby alleviating the von-Neumann bottleneck. We evaluate SPARE for two different ROM-Embedded RAM structures - CMOS based ROM-Embedded SRAMs (R-SRAMs) and STT-MRAM based ROM-Embedded MRAMs (R-MRAMs). Moreover, we analyze trade-offs in terms of energy, area and performance, for using the two technologies on a range of image classification benchmarks. Furthermore, we leverage the additional storage density to implement complex neuro-synaptic functionalities. This enhances the utility of the proposed architecture by provisioning implementation of any neuron/synaptic behavior as necessitated by the application. Our results show up-to 1.75x, 1.95x and 1.95x improvement in energy, iso-storage area, and iso-area performance, respectively, by using neural network accelerators built on ROM-embedded RAM primitives.

cs.ET↗

arXiv2018

An All-Memristor Deep Spiking Neural Computing System: A Step Towards Realizing the Low Power,Stochastic Brain

Deep 'Analog Artificial Neural Networks' (ANNs) perform complex classification problems with remarkably high accuracy. However, they rely on humongous amount of power to perform the calculations, veiling the accuracy benefits. The biological brain on the other hand is significantly more powerful than such networks and consumes orders of magnitude less power, indicating us about some conceptual mismatch. Given that the biological neurons communicate using energy efficient trains of spikes, and the behavior is non-deterministic, incorporating these effects in Deep Artificial Neural Networks may drive us few steps towards a more realistic neuron. In this work, we propose how the inherent stochasticity of nano-scale resistive devices can be harnessed to emulate the functionality of a spiking neuron that can be incorporated in deep stochastic Spiking Neural Networks (SNN). At the algorithmic level, we propose how the training can be modified to convert an ANN to an SNN while supporting the stochastic activation function offered by these devices. We devise circuit architectures to incorporate stochastic memristive neurons along with memristive crossbars which perform the functionality of the synaptic weights. We tested the proposed All Memristor deep stochastic SNN for image classification and observed only about 1% degradation in accuracy with the ANN baseline after incorporating the circuit and device related non-idealities. We witnessed that the network is robust to certain variations and consumes ~ 6.4x less energy than its CMOS counterpart.

cs.ET↗

arXiv2018

TraNNsformer: Neural network transformation for memristive crossbar based neuromorphic system design

Implementation of Neuromorphic Systems using post Complementary Metal-Oxide-Semiconductor (CMOS) technology based Memristive Crossbar Array (MCA) has emerged as a promising solution to enable low-power acceleration of neural networks. However, the recent trend to design Deep Neural Networks (DNNs) for achieving human-like cognitive abilities poses significant challenges towards the scalable design of neuromorphic systems (due to the increase in computation/storage demands). Network pruning [7] is a powerful technique to remove redundant connections for designing optimally connected (maximally sparse) DNNs. However, such pruning techniques induce irregular connections that are incoherent to the crossbar structure. Eventually they produce DNNs with highly inefficient hardware realizations (in terms of area and energy). In this work, we propose TraNNsformer - an integrated training framework that transforms DNNs to enable their efficient realization on MCA-based systems. TraNNsformer first prunes the connectivity matrix while forming clusters with the remaining connections. Subsequently, it retrains the network to fine tune the connections and reinforce the clusters. This is done iteratively to transform the original connectivity into an optimally pruned and maximally clustered mapping. Without accuracy loss, TraNNsformer reduces the area (energy) consumption by 28% - 55% (49% - 67%) with respect to the original network. Compared to network pruning, TraNNsformer achieves 28% - 49% (15% - 29%) area (energy) savings. Furthermore, TraNNsformer is a technology-aware framework that allows mapping a given DNN to any MCA size permissible by the memristive technology for reliable operations.

cs.ET↗

arXiv2017

FALCON: Feature Driven Selective Classification for Energy-Efficient Image Recognition

Machine-learning algorithms have shown outstanding image recognition or classification performance for computer vision applications. However, the compute and energy requirement for implementing such classifier models for large-scale problems is quite high. In this paper, we propose Feature Driven Selective Classification (FALCON) inspired by the biological visual attention mechanism in the brain to optimize the energy-efficiency of machine-learning classifiers. We use the consensus in the characteristic features (color/texture) across images in a dataset to decompose the original classification problem and construct a tree of classifiers (nodes) with a generic-to-specific transition in the classification hierarchy. The initial nodes of the tree separate the instances based on feature information and selectively enable the latter nodes to perform object specific classification. The proposed methodology allows selective activation of only those branches and nodes of the classification tree that are relevant to the input while keeping the remaining nodes idle. Additionally, we propose a programmable and scalable Neuromorphic Engine (NeuE) that utilizes arrays of specialized neural computational elements to execute the FALCON based classifier models for diverse datasets. The structure of FALCON facilitates the reuse of nodes while scaling up from small classification problems to larger ones thus allowing us to construct classifier implementations that are significantly more efficient. We evaluate our approach for a 12-object classification task on the Caltech101 dataset and 10-object task on CIFAR-10 dataset by constructing FALCON models on the NeuE platform in 45nm technology. Our results demonstrate significant improvement in energy-efficiency and training time for minimal loss in output quality.

cs.CV↗

arXiv2017

RESPARC: A Reconfigurable and Energy-Efficient Architecture with Memristive Crossbars for Deep Spiking Neural Networks

Neuromorphic computing using post-CMOS technologies is gaining immense popularity due to its promising abilities to address the memory and power bottlenecks in von-Neumann computing systems. In this paper, we propose RESPARC - a reconfigurable and energy efficient architecture built-on Memristive Crossbar Arrays (MCA) for deep Spiking Neural Networks (SNNs). Prior works were primarily focused on device and circuit implementations of SNNs on crossbars. RESPARC advances this by proposing a complete system for SNN acceleration and its subsequent analysis. RESPARC utilizes the energy-efficiency of MCAs for inner-product computation and realizes a hierarchical reconfigurable design to incorporate the data-flow patterns in an SNN in a scalable fashion. We evaluate the proposed architecture on different SNNs ranging in complexity from 2k-230k neurons and 1.2M-5.5M synapses. Simulation results on these networks show that compared to the baseline digital CMOS architecture, RESPARC achieves 500X (15X) efficiency in energy benefits at 300X (60X) higher throughput for multi-layer perceptrons (deep convolutional networks). Furthermore, RESPARC is a technology-aware architecture that maps a given SNN topology to the most optimized MCA size for the given crossbar technology.

cs.ET↗