Malte J. Rasch

Manuel Le Gallo, Corey Lammie, Julian Buechel et al.

Analog In-Memory Computing (AIMC) is a promising approach to reduce the latency and energy consumption of Deep Neural Network (DNN) inference and training. However, the noisy and non-linear device characteristics, and the non-ideal peripheral circuitry in AIMC chips, require adapting DNNs to be deployed on such hardware to achieve equivalent accuracy to digital computing. In this tutorial, we provide a deep dive into how such adaptations can be achieved and evaluated using the recently released IBM Analog Hardware Acceleration Kit (AIHWKit), freely available at https://github.com/IBM/aihwkit. The AIHWKit is a Python library that simulates inference and training of DNNs using AIMC. We present an in-depth description of the AIHWKit design, functionality, and best practices to properly perform inference and training. We also present an overview of the Analog AI Cloud Composer, a platform that provides the benefits of using the AIHWKit simulation in a fully managed cloud setting along with physical AIMC hardware access, freely available at https://aihw-composer.draco.res.ibm.com. Finally, we show examples on how users can expand and customize AIHWKit for their own needs. This tutorial is accompanied by comprehensive Jupyter Notebook code examples that can be run using AIHWKit, which can be downloaded from https://github.com/IBM/aihwkit/tree/master/notebooks/tutorial.

Malte J. Rasch, Charles Mackin, Manuel Le Gallo et al.

Analog in-memory computing (AIMC) -- a promising approach for energy-efficient acceleration of deep learning workloads -- computes matrix-vector multiplications (MVMs) but only approximately, due to nonidealities that often are non-deterministic or nonlinear. This can adversely impact the achievable deep neural network (DNN) inference accuracy as compared to a conventional floating point (FP) implementation. While retraining has previously been suggested to improve robustness, prior work has explored only a few DNN topologies, using disparate and overly simplified AIMC hardware models. Here, we use hardware-aware (HWA) training to systematically examine the accuracy of AIMC for multiple common artificial intelligence (AI) workloads across multiple DNN topologies, and investigate sensitivity and robustness to a broad set of nonidealities. By introducing a new and highly realistic AIMC crossbar-model, we improve significantly on earlier retraining approaches. We show that many large-scale DNNs of various topologies, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), and transformers, can in fact be successfully retrained to show iso-accuracy on AIMC. Our results further suggest that AIMC nonidealities that add noise to the inputs or outputs, not the weights, have the largest impact on DNN accuracy, and that RNNs are particularly robust to all nonidealities.

Malte J. Rasch, Fabio Carta, Omebayode Fagbohungbe et al.

In-memory computing with resistive crossbar arrays has been suggested to accelerate deep-learning workloads in highly efficient manner. To unleash the full potential of in-memory computing, it is desirable to accelerate the training as well as inference for large deep neural networks (DNNs). In the past, specialized in-memory training algorithms have been proposed that not only accelerate the forward and backward passes, but also establish tricks to update the weight in-memory and in parallel. However, the state-of-the-art algorithm (Tiki-Taka version 2 (TTv2)) still requires near perfect offset correction and suffers from potential biases that might occur due to programming and estimation inaccuracies, as well as longer-term instabilities of the device materials. Here we propose and describe two new and improved algorithms for in-memory computing (Chopped-TTv2 (c-TTv2) and Analog Gradient Accumulation with Dynamic reference (AGAD)), that retain the same runtime complexity but correct for any remaining offsets using choppers. These algorithms greatly relax the device requirements and thus expanding the scope of possible materials potentially employed for such fast in-memory DNN training.

Hadjer Benmeziane, Corey Lammie, Irem Boybat et al.

The advancement of Deep Learning (DL) is driven by efficient Deep Neural Network (DNN) design and new hardware accelerators. Current DNN design is primarily tailored for general-purpose use and deployment on commercially viable platforms. Inference at the edge requires low latency, compact and power-efficient models, and must be cost-effective. Digital processors based on typical von Neumann architectures are not conducive to edge AI given the large amounts of required data movement in and out of memory. Conversely, analog/mixed signal in-memory computing hardware accelerators can easily transcend the memory wall of von Neuman architectures when accelerating inference workloads. They offer increased area and power efficiency, which are paramount in edge resource-constrained environments. In this paper, we propose AnalogNAS, a framework for automated DNN design targeting deployment on analog In-Memory Computing (IMC) inference accelerators. We conduct extensive hardware simulations to demonstrate the performance of AnalogNAS on State-Of-The-Art (SOTA) models in terms of accuracy and deployment efficiency on various Tiny Machine Learning (TinyML) tasks. We also present experimental results that show AnalogNAS models achieving higher accuracy than SOTA models when implemented on a 64-core IMC chip based on Phase Change Memory (PCM). The AnalogNAS search code is released: https://github.com/IBM/analog-nas

Malte J. Rasch, Diego Moreda, Tayfun Gokmen et al.

We introduce the IBM Analog Hardware Acceleration Kit, a new and first of a kind open source toolkit to simulate analog crossbar arrays in a convenient fashion from within PyTorch (freely available at https://github.com/IBM/aihwkit). The toolkit is under active development and is centered around the concept of an "analog tile" which captures the computations performed on a crossbar array. Analog tiles are building blocks that can be used to extend existing network modules with analog components and compose arbitrary artificial neural networks (ANNs) using the flexibility of the PyTorch framework. Analog tiles can be conveniently configured to emulate a plethora of different analog hardware characteristics and their non-idealities, such as device-to-device and cycle-to-cycle variations, resistive device response curves, and weight and output noise. Additionally, the toolkit makes it possible to design custom unit cell configurations and to use advanced analog optimization algorithms such as Tiki-Taka. Moreover, the backward and update behavior can be set to "ideal" to enable hardware-aware training features for chips that target inference acceleration only. To evaluate the inference accuracy of such chips over time, we provide statistical programming noise and drift models calibrated on phase-change memory hardware. Our new toolkit is fully GPU accelerated and can be used to conveniently estimate the impact of material properties and non-idealities of future analog technology on the accuracy for arbitrary ANNs.

J. Luquin, C. Mackin, S. Ambrogio et al.

Analog In-Memory Compute (AIMC) can improve the energy efficiency of Deep Learning by orders of magnitude. Yet analog-domain device and circuit non-idealities -- within the analog ``Tiles'' performing Matrix-Vector Multiply (MVM) operations -- can degrade neural-network task accuracy. We quantify the impact of low-level distortions and noise, and develop a mathematical model for Multiply-ACcumulate (MAC) operations mapped to analog tiles. Instantaneous-current IR-drop (the most significant circuit non-ideality), and ADC quantization effects are fully captured by this model, which can predict MVM tile-outputs both rapidly and accurately, as compared to much slower rigorous circuit simulations. A statistical model of PCM read noise at nanosecond timescales is derived from -- and matched against -- experimental measurements. We integrate these (statistical) device and (deterministic) circuit effects into a PyTorch-based framework to assess the accuracy impact on the BERT and ALBERT Transformer networks. We show that hardware-aware fine-tuning using simple Gaussian noise provides resilience against ADC quantization and PCM read noise effects, but is less effective against IR-drop. This is because IR-drop -- although deterministic -- is non-linear, is changing significantly during the time-integration window, and is ultimately dependent on all the excitations being introduced in parallel into the analog tile. The apparent inability of simple Gaussian noise applied during training to properly prepare a DNN network for IR-drop during inference implies that more complex training approaches -- incorporating advances such as the Tile-circuit model introduced here -- will be critical for resilient deployment of large neural networks onto AIMC hardware.

Zhaoxian Wu, Tayfun Gokmen, Malte J. Rasch et al.

Given the high economic and environmental costs of using large vision or language models, analog in-memory accelerators present a promising solution for energy-efficient AI. While inference on analog accelerators has been studied recently, the training perspective is underexplored. Recent studies have shown that the "workhorse" of digital AI training - stochastic gradient descent (SGD) algorithm converges inexactly when applied to model training on non-ideal devices. This paper puts forth a theoretical foundation for gradient-based training on analog devices. We begin by characterizing the non-convergent issue of SGD, which is caused by the asymmetric updates on the analog devices. We then provide a lower bound of the asymptotic error to show that there is a fundamental performance limit of SGD-based analog training rather than an artifact of our analysis. To address this issue, we study a heuristic analog algorithm called Tiki-Taka that has recently exhibited superior empirical performance compared to SGD and rigorously show its ability to exactly converge to a critical point and hence eliminates the asymptotic error. The simulations verify the correctness of the analyses.

Hyungjun Kim, Malte Rasch, Tayfun Gokmen et al.

A resistive memory device-based computing architecture is one of the promising platforms for energy-efficient Deep Neural Network (DNN) training accelerators. The key technical challenge in realizing such accelerators is to accumulate the gradient information without a bias. Unlike the digital numbers in software which can be assigned and accessed with desired accuracy, numbers stored in resistive memory devices can only be manipulated following the physics of the device, which can significantly limit the training performance. Therefore, additional techniques and algorithm-level remedies are required to achieve the best possible performance in resistive memory device-based accelerators. In this paper, we analyze asymmetric conductance modulation characteristics in RRAM by Soft-bound synapse model and present an in-depth analysis on the relationship between device characteristics and DNN model accuracy using a 3-layer DNN trained on the MNIST dataset. We show that the imbalance between up and down update leads to a poor network performance. We introduce a concept of symmetry point and propose a zero-shifting technique which can compensate imbalance by programming the reference device and changing the zero value point of the weight. By using this zero-shifting method, we show that network performance dramatically improves for imbalanced synapse devices.

Malte J. Rasch, Tayfun Gokmen, Wilfried Haensch

Accelerating training of artificial neural networks (ANN) with analog resistive crossbar arrays is a promising idea. While the concept has been verified on very small ANNs and toy data sets (such as MNIST), more realistically sized ANNs and datasets have not yet been tackled. However, it is to be expected that device materials and hardware design constraints, such as noisy computations, finite number of resistive states of the device materials, saturating weight and activation ranges, and limited precision of analog-to-digital converters, will cause significant challenges to the successful training of state-of-the-art ANNs. By using analog hardware aware ANN training simulations, we here explore a number of simple algorithmic compensatory measures to cope with analog noise and limited weight and output ranges and resolutions, that dramatically improve the simulated training performances on RPU arrays on intermediately to large-scale ANNs.

Malte J. Rasch, Tayfun Gokmen, Mattia Rigotti et al.

Analog arrays are a promising upcoming hardware technology with the potential to drastically speed up deep learning. Their main advantage is that they compute matrix-vector products in constant time, irrespective of the size of the matrix. However, early convolution layers in ConvNets map very unfavorably onto analog arrays, because kernel matrices are typically small and the constant time operation needs to be sequentially iterated a large number of times, reducing the speed up advantage for ConvNets. Here, we propose to replicate the kernel matrix of a convolution layer on distinct analog arrays, and randomly divide parts of the compute among them, so that multiple kernel matrices are trained in parallel. With this modification, analog arrays execute ConvNets with an acceleration factor that is proportional to the number of kernel matrices used per layer (here tested 16-128). Despite having more free parameters, we show analytically and in numerical experiments that this convolution architecture is self-regularizing and implicitly learns similar filters across arrays. We also report superior performance on a number of datasets and increased robustness to adversarial attacks. Our investigation suggests to revise the notion that mixed analog-digital hardware is not suitable for ConvNets.

Tayfun Gokmen, Malte Rasch, Wilfried Haensch

In our previous work we have shown that resistive cross point devices, so called Resistive Processing Unit (RPU) devices, can provide significant power and speed benefits when training deep fully connected networks as well as convolutional neural networks. In this work, we further extend the RPU concept for training recurrent neural networks (RNNs) namely LSTMs. We show that the mapping of recurrent layers is very similar to the mapping of fully connected layers and therefore the RPU concept can potentially provide large acceleration factors for RNNs as well. In addition, we study the effect of various device imperfections and system parameters on training performance. Symmetry of updates becomes even more crucial for RNNs; already a few percent asymmetry results in an increase in the test error compared to the ideal case trained with floating point numbers. Furthermore, the input signal resolution to device arrays needs to be at least 7 bits for successful training. However, we show that a stochastic rounding scheme can reduce the input signal resolution back to 5 bits. Further, we find that RPU device variations and hardware noise are enough to mitigate overfitting, so that there is less need for using dropout. We note that the models trained here are roughly 1500 times larger than the fully connected network trained on MNIST dataset in terms of the total number of multiplication and summation operations performed per epoch. Thus, here we attempt to study the validity of the RPU approach for large scale networks.