Onur Mutlu

Rahul Bera, Konstantinos Kanellopoulos, Shankar Balachandran et al.

Long-latency load requests continue to limit the performance of high-performance processors. To increase the latency tolerance of a processor, architects have primarily relied on two key techniques: sophisticated data prefetchers and large on-chip caches. In this work, we show that: 1) even a sophisticated state-of-the-art prefetcher can only predict half of the off-chip load requests on average across a wide range of workloads, and 2) due to the increasing size and complexity of on-chip caches, a large fraction of the latency of an off-chip load request is spent accessing the on-chip cache hierarchy. The goal of this work is to accelerate off-chip load requests by removing the on-chip cache access latency from their critical path. To this end, we propose a new technique called Hermes, whose key idea is to: 1) accurately predict which load requests might go off-chip, and 2) speculatively fetch the data required by the predicted off-chip loads directly from the main memory, while also concurrently accessing the cache hierarchy for such loads. To enable Hermes, we develop a new lightweight, perceptron-based off-chip load prediction technique that learns to identify off-chip load requests using multiple program features (e.g., sequence of program counters). For every load request, the predictor observes a set of program features to predict whether or not the load would go off-chip. If the load is predicted to go off-chip, Hermes issues a speculative request directly to the memory controller once the load's physical address is generated. If the prediction is correct, the load eventually misses the cache hierarchy and waits for the ongoing speculative request to finish, thus hiding the on-chip cache hierarchy access latency from the critical path of the off-chip load. Our evaluation shows that Hermes significantly improves performance of a state-of-the-art baseline. We open-source Hermes.

Meryem Banu Cavlak, Gagandeep Singh, Mohammed Alser et al.

Basecalling is an essential step in nanopore sequencing analysis where the raw signals of nanopore sequencers are converted into nucleotide sequences, i.e., reads. State-of-the-art basecallers employ complex deep learning models to achieve high basecalling accuracy. This makes basecalling computationally inefficient and memory-hungry, bottlenecking the entire genome analysis pipeline. However, for many applications, the majority of reads do no match the reference genome of interest (i.e., target reference) and thus are discarded in later steps in the genomics pipeline, wasting the basecalling computation. To overcome this issue, we propose TargetCall, the first pre-basecalling filter to eliminate the wasted computation in basecalling. TargetCall's key idea is to discard reads that will not match the target reference (i.e., off-target reads) prior to basecalling. TargetCall consists of two main components: (1) LightCall, a lightweight neural network basecaller that produces noisy reads; and (2) Similarity Check, which labels each of these noisy reads as on-target or off-target by matching them to the target reference. Our thorough experimental evaluations show that TargetCall 1) improves the end-to-end basecalling runtime performance of the state-of-the-art basecaller by 3.31x while maintaining high (98.88%) recall in keeping on-target reads, 2) maintains high accuracy in downstream analysis, and 3) achieves better runtime performance, throughput, recall, precision, and generality compared to prior works. TargetCall is available at https://github.com/CMU-SAFARI/TargetCall.

Jeremie S. Kim, Can Firtina, Meryem Banu Cavlak et al.

AirLift is the first read remapping tool that enables users to quickly and comprehensively map a read set, that had been previously mapped to one reference genome, to another similar reference. Users can then quickly run a downstream analysis of read sets for each latest reference release. Compared to the state-of-the-art method for remapping reads (i.e., full mapping), AirLift reduces the overall execution time to remap read sets between two reference genome versions by up to 27.4x. We validate our remapping results with GATK and find that AirLift provides high accuracy in identifying ground truth SNP/INDEL variants AirLift source code and readme describing how to reproduce our results are available at https://github.com/CMU-SAFARI/AirLift.

Maurus Item, Juan Gómez-Luna, Yuxin Guo et al.

Processing-in-memory (PIM) promises to alleviate the data movement bottleneck in modern computing systems. However, current real-world PIM systems have the inherent disadvantage that their hardware is more constrained than in conventional processors (CPU, GPU), due to the difficulty and cost of building processing elements near or inside the memory. As a result, general-purpose PIM architectures support fairly limited instruction sets and struggle to execute complex operations such as transcendental functions and other hard-to-calculate operations (e.g., square root). These operations are particularly important for some modern workloads, e.g., activation functions in machine learning applications. In order to provide support for transcendental (and other hard-to-calculate) functions in general-purpose PIM systems, we present \emph{TransPimLib}, a library that provides CORDIC-based and LUT-based methods for trigonometric functions, hyperbolic functions, exponentiation, logarithm, square root, etc. We develop an implementation of TransPimLib for the UPMEM PIM architecture and perform a thorough evaluation of TransPimLib's methods in terms of performance and accuracy, using microbenchmarks and three full workloads (Blackscholes, Sigmoid, Softmax). We open-source all our code and datasets at~\url{https://github.com/CMU-SAFARI/transpimlib}.

Gagandeep Singh, Rakesh Nadig, Jisung Park et al.

Hybrid storage systems (HSS) use multiple different storage devices to provide high and scalable storage capacity at high performance. Recent research proposes various techniques that aim to accurately identify performance-critical data to place it in a "best-fit" storage device. Unfortunately, most of these techniques are rigid, which (1) limits their adaptivity to perform well for a wide range of workloads and storage device configurations, and (2) makes it difficult for designers to extend these techniques to different storage system configurations (e.g., with a different number or different types of storage devices) than the configuration they are designed for. We introduce Sibyl, the first technique that uses reinforcement learning for data placement in hybrid storage systems. Sibyl observes different features of the running workload as well as the storage devices to make system-aware data placement decisions. For every decision it makes, Sibyl receives a reward from the system that it uses to evaluate the long-term performance impact of its decision and continuously optimizes its data placement policy online. We implement Sibyl on real systems with various HSS configurations. Our results show that Sibyl provides 21.6%/19.9% performance improvement in a performance-oriented/cost-oriented HSS configuration compared to the best previous data placement technique. Our evaluation using an HSS configuration with three different storage devices shows that Sibyl outperforms the state-of-the-art data placement policy by 23.9%-48.2%, while significantly reducing the system architect's burden in designing a data placement mechanism that can simultaneously incorporate three storage devices. We show that Sibyl achieves 80% of the performance of an oracle policy that has complete knowledge of future access patterns while incurring a very modest storage overhead of only 124.4 KiB.

Mohammed Alser, Jeremie S. Kim, Nour Almadhoun Alserr et al.

Early detection and isolation of COVID-19 patients are essential for successful implementation of mitigation strategies and eventually curbing the disease spread. With a limited number of daily COVID-19 tests performed in every country, simulating the COVID-19 spread along with the potential effect of each mitigation strategy currently remains one of the most effective ways in managing the healthcare system and guiding policy-makers. We introduce COVIDHunter, a flexible and accurate COVID-19 outbreak simulation model that evaluates the current mitigation measures that are applied to a region, predicts COVID-19 statistics (the daily number of cases, hospitalizations, and deaths), and provides suggestions on what strength the upcoming mitigation measure should be. The key idea of COVIDHunter is to quantify the spread of COVID-19 in a geographical region by simulating the average number of new infections caused by an infected person considering the effect of external factors, such as environmental conditions (e.g., climate, temperature, humidity), different variants of concern, vaccination rate, and mitigation measures. Using Switzerland as a case study, COVIDHunter estimates that we are experiencing a deadly new wave that will peak on 26 January 2022, which is very similar in numbers to the wave we had in February 2020. The policy-makers have only one choice that is to increase the strength of the currently applied mitigation measures for 30 days. Unlike existing models, the COVIDHunter model accurately monitors and predicts the daily number of cases, hospitalizations, and deaths due to COVID-19. Our model is flexible to configure and simple to modify for modeling different scenarios under different environmental conditions and mitigation measures. We release the source code of the COVIDHunter implementation at https://github.com/CMU-SAFARI/COVIDHunter.

Geraldo F. Oliveira, Juan Gómez-Luna, Saugata Ghose et al.

Neural networks (NNs) are growing in importance and complexity. A neural network's performance (and energy efficiency) can be bound either by computation or memory resources. The processing-in-memory (PIM) paradigm, where computation is placed near or within memory arrays, is a viable solution to accelerate memory-bound NNs. However, PIM architectures vary in form, where different PIM approaches lead to different trade-offs. Our goal is to analyze, discuss, and contrast DRAM-based PIM architectures for NN performance and energy efficiency. To do so, we analyze three state-of-the-art PIM architectures: (1) UPMEM, which integrates processors and DRAM arrays into a single 2D chip; (2) Mensa, a 3D-stack-based PIM architecture tailored for edge devices; and (3) SIMDRAM, which uses the analog principles of DRAM to execute bit-serial operations. Our analysis reveals that PIM greatly benefits memory-bound NNs: (1) UPMEM provides 23x the performance of a high-end GPU when the GPU requires memory oversubscription for a general matrix-vector multiplication kernel; (2) Mensa improves energy efficiency and throughput by 3.0x and 3.1x over the Google Edge TPU for 24 Google edge NN models; and (3) SIMDRAM outperforms a CPU/GPU by 16.7x/1.4x for three binary NNs. We conclude that the ideal PIM architecture for NN models depends on a model's distinct attributes, due to the inherent architectural design choices.

Juan Gómez-Luna, Yuxin Guo, Sylvan Brocard et al.

Training machine learning (ML) algorithms is a computationally intensive process, which is frequently memory-bound due to repeatedly accessing large training datasets. As a result, processor-centric systems (e.g., CPU, GPU) suffer from costly data movement between memory units and processing units, which consumes large amounts of energy and execution cycles. Memory-centric computing systems, i.e., with processing-in-memory (PIM) capabilities, can alleviate this data movement bottleneck. Our goal is to understand the potential of modern general-purpose PIM architectures to accelerate ML training. To do so, we (1) implement several representative classic ML algorithms (namely, linear regression, logistic regression, decision tree, K-Means clustering) on a real-world general-purpose PIM architecture, (2) rigorously evaluate and characterize them in terms of accuracy, performance and scaling, and (3) compare to their counterpart implementations on CPU and GPU. Our evaluation on a real memory-centric computing system with more than 2500 PIM cores shows that general-purpose PIM architectures can greatly accelerate memory-bound ML workloads, when the necessary operations and datatypes are natively supported by PIM hardware. For example, our PIM implementation of decision tree is $27\times$ faster than a state-of-the-art CPU version on an 8-core Intel Xeon, and $1.34\times$ faster than a state-of-the-art GPU version on an NVIDIA A100. Our K-Means clustering on PIM is $2.8\times$ and $3.2\times$ than state-of-the-art CPU and GPU versions, respectively. To our knowledge, our work is the first one to evaluate ML training on a real-world PIM architecture. We conclude with key observations, takeaways, and recommendations that can inspire users of ML workloads, programmers of PIM architectures, and hardware designers & architects of future memory-centric computing systems.

Geraldo F. Oliveira, Amirali Boroumand, Saugata Ghose et al.

Today's computing systems require moving data back-and-forth between computing resources (e.g., CPUs, GPUs, accelerators) and off-chip main memory so that computation can take place on the data. Unfortunately, this data movement is a major bottleneck for system performance and energy consumption. One promising execution paradigm that alleviates the data movement bottleneck in modern and emerging applications is processing-in-memory (PIM), where the cost of data movement to/from main memory is reduced by placing computation capabilities close to memory. Naively employing PIM to accelerate data-intensive workloads can lead to sub-optimal performance due to the many design constraints PIM substrates impose. Therefore, many recent works co-design specialized PIM accelerators and algorithms to improve performance and reduce the energy consumption of (i) applications from various application domains; and (ii) various computing environments, including cloud systems, mobile systems, and edge devices. We showcase the benefits of co-designing algorithms and hardware in a way that efficiently takes advantage of the PIM paradigm for two modern data-intensive applications: (1) machine learning inference models for edge devices and (2) hybrid transactional/analytical processing databases for cloud systems. We follow a two-step approach in our system design. In the first step, we extensively analyze the computation and memory access patterns of each application to gain insights into its hardware/software requirements and major sources of performance and energy bottlenecks in processor-centric systems. In the second step, we leverage the insights from the first step to co-design algorithms and hardware accelerators to enable high-performance and energy-efficient data-centric architectures for each application.

André Santos, João Dinis Ferreira, Onur Mutlu et al.

In recent years, Convolutional Neural Networks (CNNs) have become the standard class of deep neural network for image processing, classification and segmentation tasks. However, the large strides in accuracy obtained by CNNs have been derived from increasing the complexity of network topologies, which incurs sizeable performance and energy penalties in the training and inference of CNNs. Many recent works have validated the effectiveness of parameter quantization, which consists in reducing the bit width of the network's parameters, to enable the attainment of considerable performance and energy efficiency gains without significantly compromising accuracy. However, it is difficult to compare the relative effectiveness of different quantization methods. To address this problem, we introduce RedBit, an open-source framework that provides a transparent, extensible and easy-to-use interface to evaluate the effectiveness of different algorithms and parameter configurations on network accuracy. We use RedBit to perform a comprehensive survey of five state-of-the-art quantization methods applied to the MNIST, CIFAR-10 and ImageNet datasets. We evaluate a total of 2300 individual bit width combinations, independently tuning the width of the network's weight and input activation parameters, from 32 bits down to 1 bit (e.g., 8/8, 2/2, 1/32, 1/1, for weights/activations). Upwards of 20000 hours of computing time in a pool of state-of-the-art GPUs were used to generate all the results in this paper. For 1-bit quantization, the accuracy losses for the MNIST, CIFAR-10 and ImageNet datasets range between [0.26%, 0.79%], [9.74%, 32.96%] and [10.86%, 47.36%] top-1, respectively. We actively encourage the reader to download the source code and experiment with RedBit, and to submit their own observed results to our public repository, available at https://github.com/IT-Coimbra/RedBit.

Aditya Manglik, Minesh Patel, Haiyu Mao et al.

Resistive Random-Access Memory (RRAM) is well-suited to accelerate neural network (NN) workloads as RRAM-based Processing-in-Memory (PIM) architectures natively support highly-parallel multiply-accumulate (MAC) operations that form the backbone of most NN workloads. Unfortunately, NN workloads such as transformers require support for non-MAC operations (e.g., softmax) that RRAM cannot provide natively. Consequently, state-of-the-art works either integrate additional digital logic circuits to support the non-MAC operations or offload the non-MAC operations to CPU/GPU, resulting in significant performance and energy efficiency overheads due to data movement. In this work, we propose NEON, a novel compiler optimization to enable the end-to-end execution of the NN workload in RRAM. The key idea of NEON is to transform each non-MAC operation into a lightweight yet highly-accurate neural network. Utilizing neural networks to approximate the non-MAC operations provides two advantages: 1) We can exploit the key strength of RRAM, i.e., highly-parallel MAC operation, to flexibly and efficiently execute non-MAC operations in memory. 2) We can simplify RRAM's microarchitecture by eliminating the additional digital logic circuits while reducing the data movement overheads. Acceleration of the non-MAC operations in memory enables NEON to achieve a 2.28x speedup compared to an idealized digital logic-based RRAM. We analyze the trade-offs associated with the transformation and demonstrate feasible use cases for NEON across different substrates.

Can Firtina, Kamlesh Pillai, Gurpreet S. Kalsi et al.

Profile hidden Markov models (pHMMs) are widely employed in various bioinformatics applications to identify similarities between biological sequences, such as DNA or protein sequences. In pHMMs, sequences are represented as graph structures. These probabilities are subsequently used to compute the similarity score between a sequence and a pHMM graph. The Baum-Welch algorithm, a prevalent and highly accurate method, utilizes these probabilities to optimize and compute similarity scores. However, the Baum-Welch algorithm is computationally intensive, and existing solutions offer either software-only or hardware-only approaches with fixed pHMM designs. We identify an urgent need for a flexible, high-performance, and energy-efficient HW/SW co-design to address the major inefficiencies in the Baum-Welch algorithm for pHMMs. We introduce ApHMM, the first flexible acceleration framework designed to significantly reduce both computational and energy overheads associated with the Baum-Welch algorithm for pHMMs. ApHMM tackles the major inefficiencies in the Baum-Welch algorithm by 1) designing flexible hardware to accommodate various pHMM designs, 2) exploiting predictable data dependency patterns through on-chip memory with memoization techniques, 3) rapidly filtering out negligible computations using a hardware-based filter, and 4) minimizing redundant computations. ApHMM achieves substantial speedups of 15.55x - 260.03x, 1.83x - 5.34x, and 27.97x when compared to CPU, GPU, and FPGA implementations of the Baum-Welch algorithm, respectively. ApHMM outperforms state-of-the-art CPU implementations in three key bioinformatics applications: 1) error correction, 2) protein family search, and 3) multiple sequence alignment, by 1.29x - 59.94x, 1.03x - 1.75x, and 1.03x - 1.95x, respectively, while improving their energy efficiency by 64.24x - 115.46x, 1.75x, 1.96x.

Gagandeep Singh, Dionysios Diamantopoulos, Juan Gómez-Luna et al.

Machine learning has recently gained traction as a way to overcome the slow accelerator generation and implementation process on an FPGA. It can be used to build performance and resource usage models that enable fast early-stage design space exploration. First, training requires large amounts of data (features extracted from design synthesis and implementation tools), which is cost-inefficient because of the time-consuming accelerator design and implementation process. Second, a model trained for a specific environment cannot predict performance or resource usage for a new, unknown environment. In a cloud system, renting a platform for data collection to build an ML model can significantly increase the total-cost-ownership (TCO) of a system. Third, ML-based models trained using a limited number of samples are prone to overfitting. To overcome these limitations, we propose LEAPER, a transfer learning-based approach for prediction of performance and resource usage in FPGA-based systems. The key idea of LEAPER is to transfer an ML-based performance and resource usage model trained for a low-end edge environment to a new, high-end cloud environment to provide fast and accurate predictions for accelerator implementation. Experimental results show that LEAPER (1) provides, on average across six workloads and five FPGAs, 85% accuracy when we use our transferred model for prediction in a cloud environment with 5-shot learning and (2) reduces design-space exploration time for accelerator implementation on an FPGA by 10x, from days to only a few hours.

Juan Gómez-Luna, Yuxin Guo, Sylvan Brocard et al.

Training machine learning algorithms is a computationally intensive process, which is frequently memory-bound due to repeatedly accessing large training datasets. As a result, processor-centric systems (e.g., CPU, GPU) suffer from costly data movement between memory units and processing units, which consumes large amounts of energy and execution cycles. Memory-centric computing systems, i.e., computing systems with processing-in-memory (PIM) capabilities, can alleviate this data movement bottleneck. Our goal is to understand the potential of modern general-purpose PIM architectures to accelerate machine learning training. To do so, we (1) implement several representative classic machine learning algorithms (namely, linear regression, logistic regression, decision tree, K-means clustering) on a real-world general-purpose PIM architecture, (2) characterize them in terms of accuracy, performance and scaling, and (3) compare to their counterpart implementations on CPU and GPU. Our experimental evaluation on a memory-centric computing system with more than 2500 PIM cores shows that general-purpose PIM architectures can greatly accelerate memory-bound machine learning workloads, when the necessary operations and datatypes are natively supported by PIM hardware. To our knowledge, our work is the first one to evaluate training of machine learning algorithms on a real-world general-purpose PIM architecture.

William Andrew Simon, Leonid Yavits, Konstantina Koliogeorgi et al.

Low-cost, high-throughput DNA and RNA sequencing (HTS) data is the backbone of the life sciences. Genome sequencing is now becoming a part of Predictive, Preventive, Personalized, and Participatory (termed 'P4') medicine. All genomic data are currently processed in energy-hungry computer clusters and centers, necessitating data transfer, consuming substantial energy, and wasting valuable time. Therefore, there is a need for fast, energy-efficient, and cost-efficient technologies that enable genomics research without requiring data centers and cloud platforms. We recently launched the BioPIM Project to leverage emerging processing-in-memory (PIM) technologies to enable energy- and cost-efficient analysis of bioinformatics workloads. The BioPIM Project focuses on co-designing algorithms and data structures commonly used in genomics with several PIM architectures to achieve the highest cost, energy, and time savings.

F. Nisa Bostanci, Haocong Luo, Ataberk Olgun et al.

A MICRO 2024 best paper runner-up publication (the Mess paper) with all three artifact badges awarded (including ``Reproducible'') proposes a new benchmark to evaluate real and simulated memory system performance. The publication contends that Ramulator 2.0 and DAMOV (ZSim+Ramulator) (along with other existing memory system simulators) ``poorly resemble the actual system performance'' and asserts that their simulator is better. In this paper, we show that the Mess paper has 1) demonstrable technical misconfigurations, 2) methodological errors in interpreting simulation statistics, and 3) an incomplete artifact that makes its key results irreproducible. We demonstrate that the Ramulator 2.0 simulation results reported in the Mess paper are incorrect due to multiple configuration errors instead of inherent simulation inaccuracy claimed by the Mess paper. We show that by correctly configuring Ramulator 2.0, Ramulator 2.0's simulated memory system performance actually resembles real system characteristics well, and thus a key claimed contribution of the Mess paper is factually incorrect. We also identify that the DAMOV simulation results in the Mess paper use wrong simulation statistics that are unrelated to the simulated DRAM performance. We show that DAMOV's simulated DRAM latency is not constant, in contrast to the Mess paper's claim. Moreover, the Mess paper's artifact repository lacks the necessary sources to fully reproduce all the Mess paper's results. We find that the experiment scripts use simulator executables and other resources that are neither described in the Mess paper nor found in the artifact repository. We strongly encourage the computer architecture community to consider our corrections to the Ramulator 2.0 and DAMOV results of the Mess paper to prevent the propagation of inaccurate and misleading results and to maintain the reliability of the scientific record.

Lorenzo Asquini, Manos Frouzakis, Juan Gómez-Luna et al.

Triangle Counting (TC) is a procedure that involves enumerating the number of triangles within a graph. It has important applications in numerous fields, such as social or biological network analysis and network security. TC is a memory-bound workload that does not scale efficiently in conventional processor-centric systems due to several memory accesses across large memory regions and low data reuse. However, recent Processing-in-Memory (PIM) architectures present a promising solution to alleviate these bottlenecks. Our work presents the first TC algorithm that leverages the capabilities of the UPMEM system, the first commercially available PIM architecture, while at the same time addressing its limitations. We use a vertex coloring technique to avoid expensive communication between PIM cores and employ reservoir sampling to address the limited amount of memory available in the PIM cores' DRAM banks. In addition, our work makes use of the Misra-Gries summary to speed up counting triangles on graphs with high-degree nodes and uniform sampling of the graph edges for quicker approximate results. Our PIM implementation surpasses state-of-the-art CPU-based TC implementations when processing dynamic graphs in Coordinate List format, showcasing the effectiveness of the UPMEM architecture in addressing TC's memory-bound challenges.

Steve Rhyner, Haocong Luo, Juan Gómez-Luna et al.

Modern Machine Learning (ML) training on large-scale datasets is a very time-consuming workload. It relies on the optimization algorithm Stochastic Gradient Descent (SGD) due to its effectiveness, simplicity, and generalization performance. Processor-centric architectures (e.g., CPUs, GPUs) commonly used for modern ML training workloads based on SGD are bottlenecked by data movement between the processor and memory units due to the poor data locality in accessing large datasets. As a result, processor-centric architectures suffer from low performance and high energy consumption while executing ML training workloads. Processing-In-Memory (PIM) is a promising solution to alleviate the data movement bottleneck by placing the computation mechanisms inside or near memory. Our goal is to understand the capabilities of popular distributed SGD algorithms on real-world PIM systems to accelerate data-intensive ML training workloads. To this end, we 1) implement several representative centralized parallel SGD algorithms on the real-world UPMEM PIM system, 2) rigorously evaluate these algorithms for ML training on large-scale datasets in terms of performance, accuracy, and scalability, 3) compare to conventional CPU and GPU baselines, and 4) discuss implications for future PIM hardware and highlight the need for a shift to an algorithm-hardware codesign. Our results demonstrate three major findings: 1) The UPMEM PIM system can be a viable alternative to state-of-the-art CPUs and GPUs for many memory-bound ML training workloads, especially when operations and datatypes are natively supported by PIM hardware, 2) it is important to carefully choose the optimization algorithms that best fit PIM, and 3) the UPMEM PIM system does not scale approximately linearly with the number of nodes for many data-intensive ML training workloads. We open source all our code to facilitate future research.

Nicola Barcarolo, Brahmaiah Gandham, Mohammad Sadrosadati et al.

Cryptographic algorithms such as AES-128 and SHA-256 are fundamental to ensuring data security and integrity. Although these algorithms are computationally efficient, their performance is often constrained by the processor-centric architectures (e.g., CPUs, GPUs), primarily due to the memory bottleneck. This constraint leads to increased latency and higher energy consumption, particularly when handling large volumes of data. To overcome these challenges, Processing-in-Memory (PIM) has emerged as a promising architectural paradigm, allowing computation to occur directly within or near memory units. By minimizing data movement between the processor and memory units, PIM can significantly accelerate cryptographic algorithms while improving energy efficiency. Several pieces of prior work have demonstrated the effectiveness of PIM at fundamentally accelerating cryptographic algorithms. However, none of the prior works have extensively demonstrated the potential of a real-world PIM system. In this paper, we want to investigate the potential and limitations of real-world PIM in accelerating cryptographic algorithms. As part of our methodology, the UPMEM PIM architecture is used to assess the scalability of cryptographic algorithms. When these algorithms operate on a single rank, their performance remains below that of modern CPUs. However, distributing the computation across multiple ranks significantly enhances performance. When all available ranks are utilized, real-world PIM can accelerate cryptographic algorithms more effectively.

Rahul Bera, Rakesh Nadig, Onur Mutlu

Despite the data-rich environment in which memory systems of modern computing platforms operate, many state-of-the-art architectural policies employed in the memory system rely on static, human-designed heuristics that fail to truly adapt to the workload and system behavior via principled learning methodologies. In this article, we propose a fundamentally different design approach: using lightweight and practical machine learning (ML) methods to enable adaptive, data-driven control throughout the memory hierarchy. We present three ML-guided architectural policies: (1) Pythia, a reinforcement learning-based data prefetcher for on-chip caches, (2) Hermes, a perceptron learning-based off-chip predictor for multi-level cache hierarchies, and (3) Sibyl, a reinforcement learning-based data placement policy for hybrid storage systems. Our evaluation shows that Pythia, Hermes, and Sibyl significantly outperform the best-prior human-designed policies, while incurring modest hardware overheads. Collectively, this article demonstrates that integrating adaptive learning into memory subsystems can lead to intelligent, self-optimizing architectures that unlock performance and efficiency gains beyond what is possible with traditional human-designed approaches.

Christina Giannoula, Peiming Yang, Ivan Fernandez et al.

Graph Neural Networks (GNNs) are emerging ML models to analyze graph-structure data. Graph Neural Network (GNN) execution involves both compute-intensive and memory-intensive kernels, the latter dominates the total time, being significantly bottlenecked by data movement between memory and processors. Processing-In-Memory (PIM) systems can alleviate this data movement bottleneck by placing simple processors near or inside to memory arrays. In this work, we introduce PyGim, an efficient ML library that accelerates GNNs on real PIM systems. We propose intelligent parallelization techniques for memory-intensive kernels of GNNs tailored for real PIM systems, and develop handy Python API for them. We provide hybrid GNN execution, in which the compute-intensive and memory-intensive kernels are executed in processor-centric and memory-centric computing systems, respectively. We extensively evaluate PyGim on a real-world PIM system with 1992 PIM cores using emerging GNN models, and demonstrate that it outperforms its state-of-the-art CPU counterpart on Intel Xeon by on average 3.04x, and achieves higher resource utilization than CPU and GPU systems. Our work provides useful recommendations for software, system and hardware designers. PyGim is publicly available at https://github.com/CMU-SAFARI/PyGim.

Harshita Gupta, Mayank Kabra, Jaewoo Park et al.

Homomorphic encryption (HE) enables computation over encrypted data, offering strong privacy guarantees for untrusted computing environments. Practical adoption remains limited by high computational complexity, large ciphertext sizes, and substantial data movement. Processor-centric architectures (CPUs, GPUs, ASICs) hit fundamental bottlenecks on HE workloads because ciphertexts are large, data locality is low, and primitives such as relinearization and bootstrapping repeatedly access large auxiliary metadata. Processing-In-Memory (PIM) is a promising mitigation by computing near or inside memory. Prior PIM proposals for HE either do not target real-world PIM systems or cover only a narrow set of operations. We comprehensively characterize HE operations on a real-world, general-purpose PIM system. We implement a complete set of HE kernels used by emerging applications (databases, machine learning) on the UPMEM PIM system, evaluate performance and scalability, compare against CPU and GPU baselines, and discuss implications for future PIM hardware. Our results demonstrate four major findings. (1) HE-based applications expose distinct bottlenecks across execution stages: some kernels are compute-bound due to modular arithmetic, while others are memory-bound due to large ciphertexts and intermediate data. These bottlenecks are exacerbated by limited per-core compute and per-bank capacity, which force frequent data movement. (2) The dominant compute bottleneck is the lack of native 64-bit modular integer multiplication, a key HE primitive. (3) Limited per-bank memory capacity is the second major bottleneck, since HE ciphertexts and auxiliary metadata do not fit and require inter-bank movement. (4) Despite these limits, PIM can be a viable alternative to state-of-the-art CPU and GPU systems for HE when equipped with native modular multiplication and efficient inter-PIM data movement.

Rakesh Nadig, Vamanan Arulchelvan, Rahul Bera et al.

Hybrid storage systems (HSS) integrate multiple storage devices with diverse characteristics to deliver high performance and capacity at low cost. The performance of an HSS highly depends on the effectiveness of two key policies: (1) the data-placement policy, which determines the best-fit storage device for incoming data, and (2) the data-migration policy, which dynamically rearranges stored data (i.e., prefetches hot data and evicts cold data) across the devices to sustain high HSS performance. Prior works optimize either data placement or data migration in isolation, which leads to suboptimal HSS performance. Unfortunately, no prior work tries to optimize both policies together. Our goal is to design a holistic data-management technique that optimizes both data-placement and data-migration policies to fully exploit the potential of an HSS, and thus significantly improve system performance. We propose Harmonia, a multi-agent reinforcement learning (RL)-based data-management technique that employs two lightweight autonomous RL agents, a data-placement agent and a data-migration agent, that adapt their policies for the current workload and HSS configuration while coordinating with each other to improve overall HSS performance. We evaluate Harmonia on real HSS configurations with up to four heterogeneous storage devices and seventeen data-intensive workloads. On performance-optimized (cost-optimized) HSS with two storage devices, Harmonia outperforms the best-performing prior approach by 49.5% (31.7%) on average. On an HSS with three (four) devices, Harmonia outperforms the best-performing prior work by 37.0% (42.0%) on average. Harmonia's performance benefits come with low latency (240ns for inference) and storage overheads (206 KiB in DRAM for both RL agents combined). We will open-source Harmonia's implementation to aid future research on HSS.

Rahul Bera, Konstantinos Kanellopoulos, Anant V. Nori et al.

Past research has proposed numerous hardware prefetching techniques, most of which rely on exploiting one specific type of program context information (e.g., program counter, cacheline address) to predict future memory accesses. These techniques either completely neglect a prefetcher's undesirable effects (e.g., memory bandwidth usage) on the overall system, or incorporate system-level feedback as an afterthought to a system-unaware prefetch algorithm. We show that prior prefetchers often lose their performance benefit over a wide range of workloads and system configurations due to their inherent inability to take multiple different types of program context and system-level feedback information into account while prefetching. In this paper, we make a case for designing a holistic prefetch algorithm that learns to prefetch using multiple different types of program context and system-level feedback information inherent to its design. To this end, we propose Pythia, which formulates the prefetcher as a reinforcement learning agent. For every demand request, Pythia observes multiple different types of program context information to make a prefetch decision. For every prefetch decision, Pythia receives a numerical reward that evaluates prefetch quality under the current memory bandwidth usage. Pythia uses this reward to reinforce the correlation between program context information and prefetch decision to generate highly accurate, timely, and system-aware prefetch requests in the future. Our extensive evaluations using simulation and hardware synthesis show that Pythia outperforms multiple state-of-the-art prefetchers over a wide range of workloads and system configurations, while incurring only 1.03% area overhead over a desktop-class processor and no software changes in workloads. The source code of Pythia can be freely downloaded from https://github.com/CMU-SAFARI/Pythia.

Kailash Gogineni, Sai Santosh Dayapule, Juan Gómez-Luna et al.

Reinforcement Learning (RL) trains agents to learn optimal behavior by maximizing reward signals from experience datasets. However, RL training often faces memory limitations, leading to execution latencies and prolonged training times. To overcome this, SwiftRL explores Processing-In-Memory (PIM) architectures to accelerate RL workloads. We achieve near-linear performance scaling by implementing RL algorithms like Tabular Q-learning and SARSA on UPMEM PIM systems and optimizing for hardware. Our experiments on OpenAI GYM environments using UPMEM hardware demonstrate superior performance compared to CPU and GPU implementations.

Yintao He, Haiyu Mao, Christina Giannoula et al.

Large language models (LLMs) are widely used for natural language understanding and text generation. An LLM model relies on a time-consuming step called LLM decoding to generate output tokens. Several prior works focus on improving the performance of LLM decoding using parallelism techniques, such as batching and speculative decoding. State-of-the-art LLM decoding has both compute-bound and memory-bound kernels. Some prior works statically identify and map these different kernels to a heterogeneous architecture consisting of both processing-in-memory (PIM) units and computation-centric accelerators. We observe that characteristics of LLM decoding kernels (e.g., whether or not a kernel is memory-bound) can change dynamically due to parameter changes to meet user and/or system demands, making (1) static kernel mapping to PIM units and computation-centric accelerators suboptimal, and (2) one-size-fits-all approach of designing PIM units inefficient due to a large degree of heterogeneity even in memory-bound kernels. In this paper, we aim to accelerate LLM decoding while considering the dynamically changing characteristics of the kernels involved. We propose PAPI (PArallel Decoding with PIM), a PIM-enabled heterogeneous architecture that exploits dynamic scheduling of compute-bound or memory-bound kernels to suitable hardware units. PAPI has two key mechanisms: (1) online kernel characterization to dynamically schedule kernels to the most suitable hardware units at runtime and (2) a PIM-enabled heterogeneous computing system that harmoniously orchestrates both computation-centric processing units and hybrid PIM units with different computing capabilities. Our experimental results on three broadly-used LLMs show that PAPI achieves 1.8$\times$ and 11.1$\times$ speedups over a state-of-the-art heterogeneous LLM accelerator and a state-of-the-art PIM-only LLM accelerator, respectively.

Shai Bergman, Zhe Yang, Julien Eudine et al.

Modern distributed file systems rely on uncoordinated, per node page caches that replicate hot data locally across the cluster. While ensuring fast local access, this architecture underutilizes aggregate cluster DRAM capacity through massive data redundancy and incurs prohibitive coherence overhead via heavyweight, lock-based protocols. In this paper, we focus on the design of a distributed page cache that treats the entire cluster's main memory as a single cache budget while preserving standard file-system interfaces and semantics. We present Distributed Page Cache (DPC), an OS-level, distributed page cache built on top of Compute Express Link (CXL) 3.0 memory semantics. DPC enforces a single-copy invariant at page granularity: each file page has exactly one owner node holding the sole resident DRAM copy, and other nodes access it via CXL-based remote mappings rather than creating replicas of the page. DPC is implemented end-to-end on a CXL-based emulation framework that models multi-host CXL 3.0 memory fabrics, enabling detailed evaluation in the absence of widespread hardware. Across real-world and representative data-sharing workloads, DPC delivers speedups of up to 12.4X, with a geometric-mean speedup of 5.6X.

Ataberk Olgun, F. Nisa Bostanci, Ismail Emir Yuksel et al.

State-of-the-art DRAM read disturbance mitigations rely on the read disturbance threshold (RDT) (e.g., the number of aggressor row activations needed to induce the first read disturbance bitflip) to securely and performance- and energy-efficiently prevent read disturbance bitflips. However, accurately and exhaustively characterizing the RDT of every DRAM row in a chip is time intensive. Rapidly determining RDT is important for enabling secure, performance- and energy-efficient systems. Our goal is to develop and evaluate a reliable and rapid read disturbance testing methodology. To that end, we develop DiscoRD building on the key results of an extensive experimental characterization study using 212 real DDR4 chips whereby we measure the RDT of hundreds of thousands of DRAM rows millions of times. We develop an empirical model for read disturbance bitflips and evaluate the probability of read-disturbance-induced uncorrectable errors when a read disturbance mechanism is configured using a single $RDT_{min}$ measurement. Using this model we demonstrate that 1) relying on a lightweight error-correcting code (ECC) alone yields relatively high uncorrectable error probability and 2) combining ECC, infrequent memory scrubbing, and configurable read disturbance mitigation mechanisms can greatly reduce the error probability. Building on our observations and analyses, we discuss the RDT of each individual row can be identified more precisely. Our results show that error tolerance, memory scrubbing, online profiling, and run-time configurable read disturbance mitigation techniques are important to enable secure and energy-efficient spatial-variation aware read disturbance mitigations. We hope that DiscoRD drives research that enables us to quantitatively navigate the performance/cost - reliability tradeoff space for read disturbance mitigation techniques.

Peiming Yang, Sankeerth Durvasula, Ivan Fernandez et al.

Processing-In-Memory (PIM) devices integrated with high-performance Host processors (e.g., GPUs) can accelerate memory-intensive kernels in Machine Learning (ML) models, including Large Language Models (LLMs), by leveraging high memory bandwidth at PIM cores. However, Host processors and PIM cores require different data layouts: Hosts need consecutive elements distributed across DRAM banks, while PIM cores need them within local banks. This necessitates data rearrangements in ML kernel execution that pose significant performance and programmability challenges, further exacerbated by the need to support diverse PIM backends. Current compilation approaches lack systematic optimization for diverse ML kernels across multiple PIM backends and may largely ignore data rearrangements during compute code optimization. We demonstrate that data rearrangements and compute code optimization are interdependent, and need to be jointly optimized during the tuning process. To address this, we design DCC, the first data-centric ML compiler for PIM systems that jointly co-optimizes data rearrangements and compute code in a unified tuning process. DCC integrates a multi-layer PIM abstraction that enables various data distribution and processing strategies on different PIM backends. DCC enables effective co-optimization by mapping data partitioning strategies to compute loop partitions, applying PIM-specific code optimizations and leveraging a fast and accurate performance prediction model to select optimal configurations. Our evaluations in various individual ML kernels demonstrate that DCC achieves up to 7.68x speedup (2.7x average) on HBM-PIM and up to 13.17x speedup (5.75x average) on AttAcc PIM backend over GPU-only execution. In end-to-end LLM inference, DCC on AttAcc accelerates GPT-3 and LLaMA-2 by up to 7.71x (4.88x average) over GPU.

Kangqi Chen, Andreas Kosmas Kakolyris, Rakesh Nadig et al.

Large Language Models (LLMs) face an inherent challenge: their knowledge is confined to the data that they have been trained on. To overcome this issue, Retrieval-Augmented Generation (RAG) complements the static training-derived knowledge of LLMs with an external knowledge repository. RAG consists of three stages: indexing, retrieval, and generation. The retrieval stage of RAG becomes a significant bottleneck in inference pipelines. In this stage, a user query is mapped to an embedding vector and an Approximate Nearest Neighbor Search (ANNS) algorithm searches for similar vectors in the database to identify relevant items. Due to the large database sizes, ANNS incurs significant data movement overheads between the host and the storage system. To alleviate these overheads, prior works propose In-Storage Processing (ISP) techniques that accelerate ANNS by performing computations inside storage. However, existing works that leverage ISP for ANNS (i) employ algorithms that are not tailored to ISP systems, (ii) do not accelerate data retrieval operations for data selected by ANNS, and (iii) introduce significant hardware modifications, limiting performance and hindering their adoption. We propose REIS, the first ISP system tailored for RAG that addresses these limitations with three key mechanisms. First, REIS employs a database layout that links database embedding vectors to their associated documents, enabling efficient retrieval. Second, it enables efficient ANNS by introducing an ISP-tailored data placement technique that distributes embeddings across the planes of the storage system and employs a lightweight Flash Translation Layer. Third, REIS leverages an ANNS engine that uses the existing computational resources inside the storage system. Compared to a server-grade system, REIS improves the performance (energy efficiency) of retrieval by an average of 13x (55x).

Maciej Besta, Florim Memedi, Zhenyu Zhang et al.

The field of natural language processing (NLP) has witnessed significant progress in recent years, with a notable focus on improving large language models' (LLM) performance through innovative prompting techniques. Among these, prompt engineering coupled with structures has emerged as a promising paradigm, with designs such as Chain-of-Thought, Tree of Thoughts, or Graph of Thoughts, in which the overall LLM reasoning is guided by a structure such as a graph. As illustrated with numerous examples, this paradigm significantly enhances the LLM's capability to solve numerous tasks, ranging from logical or mathematical reasoning to planning or creative writing. To facilitate the understanding of this growing field and pave the way for future developments, we devise a general blueprint for effective and efficient LLM reasoning schemes. For this, we conduct an in-depth analysis of the prompt execution pipeline, clarifying and clearly defining different concepts. We then build the first taxonomy of structure-enhanced LLM reasoning schemes. We focus on identifying fundamental classes of harnessed structures, and we analyze the representations of these structures, algorithms executed with these structures, and many others. We refer to these structures as reasoning topologies, because their representation becomes to a degree spatial, as they are contained within the LLM context. Our study compares existing prompting schemes using the proposed taxonomy, discussing how certain design choices lead to different patterns in performance and cost. We also outline theoretical underpinnings, relationships between prompting and other parts of the LLM ecosystem such as knowledge bases, and the associated research challenges. Our work will help to advance future prompt engineering techniques.

Jisung Park, Jeoggyun Kim, Yeseong Kim et al.

Data reduction in storage systems is becoming increasingly important as an effective solution to minimize the management cost of a data center. To maximize data-reduction efficiency, existing post-deduplication delta-compression techniques perform delta compression along with traditional data deduplication and lossless compression. Unfortunately, we observe that existing techniques achieve significantly lower data-reduction ratios than the optimal due to their limited accuracy in identifying similar data blocks. In this paper, we propose DeepSketch, a new reference search technique for post-deduplication delta compression that leverages the learning-to-hash method to achieve higher accuracy in reference search for delta compression, thereby improving data-reduction efficiency. DeepSketch uses a deep neural network to extract a data block's sketch, i.e., to create an approximate data signature of the block that can preserve similarity with other blocks. Our evaluation using eleven real-world workloads shows that DeepSketch improves the data-reduction ratio by up to 33% (21% on average) over a state-of-the-art post-deduplication delta-compression technique.

Lois Orosa, Skanda Koppula, Yaman Umuroglu et al.

Dilated and transposed convolutions are widely used in modern convolutional neural networks (CNNs). These kernels are used extensively during CNN training and inference of applications such as image segmentation and high-resolution image generation. Although these kernels have grown in popularity, they stress current compute systems due to their high memory intensity, exascale compute demands, and large energy consumption. We find that commonly-used low-power CNN inference accelerators based on spatial architectures are not optimized for both of these convolutional kernels. Dilated and transposed convolutions introduce significant zero padding when mapped to the underlying spatial architecture, significantly degrading performance and energy efficiency. Existing approaches that address this issue require significant design changes to the otherwise simple, efficient, and well-adopted architectures used to compute direct convolutions. To address this challenge, we propose EcoFlow, a new set of dataflows and mapping algorithms for dilated and transposed convolutions. These algorithms are tailored to execute efficiently on existing low-cost, small-scale spatial architectures and requires minimal changes to the network-on-chip of existing accelerators. EcoFlow eliminates zero padding through careful dataflow orchestration and data mapping tailored to the spatial architecture. EcoFlow enables flexible and high-performance transpose and dilated convolutions on architectures that are otherwise optimized for CNN inference. We evaluate the efficiency of EcoFlow on CNN training workloads and Generative Adversarial Network (GAN) training workloads. Experiments in our new cycle-accurate simulator show that EcoFlow 1) reduces end-to-end CNN training time between 7-85%, and 2) improves end-to-end GAN training performance between 29-42%, compared to state-of-the-art CNN inference accelerators.

Nour Almadhoun Alserr, Ozgur Ulusoy, Erman Ayday et al.

Motivation: Cutting the cost of DNA sequencing technology led to a quantum leap in the availability of genomic data. While sharing genomic data across researchers is an essential driver of advances in health and biomedical research, the sharing process is often infeasible due to data privacy concerns. Differential privacy is one of the rigorous mechanisms utilized to facilitate the sharing of aggregate statistics from genomic datasets without disclosing any private individual-level data. However, differential privacy can still divulge sensitive information about the dataset participants due to the correlation between dataset tuples. Results: Here, we propose GenShare model built upon Laplace-perturbation-mechanism-based DP to introduce a privacy-preserving query-answering sharing model for statistical genomic datasets that include dependency due to the inherent correlations between genomes of individuals (i.e., family ties). We demonstrate our privacy improvement over the state-of-the-art approaches for a range of practical queries including cohort discovery, minor allele frequency, and chi^2 association tests. With a fine-grained analysis of sensitivity in the Laplace perturbation mechanism and considering joint distributions, GenShare results near-achieve the formal privacy guarantees permitted by the theory of differential privacy as the queries that computed over independent tuples (only up to 6% differences). GenShare ensures that query results are as accurate as theoretically guaranteed by differential privacy. For empowering the advances in different scientific and medical research areas, GenShare presents a path toward an interactive genomic data sharing system when the datasets include participants with familial relationships.

Hasan Hassan, Yahya Can Tugrul, Jeremie S. Kim et al.

The RowHammer vulnerability in DRAM is a critical threat to system security. To protect against RowHammer, vendors commit to security-through-obscurity: modern DRAM chips rely on undocumented, proprietary, on-die mitigations, commonly known as Target Row Refresh (TRR). At a high level, TRR detects and refreshes potential RowHammer-victim rows, but its exact implementations are not openly disclosed. Security guarantees of TRR mechanisms cannot be easily studied due to their proprietary nature. To assess the security guarantees of recent DRAM chips, we present Uncovering TRR (U-TRR), an experimental methodology to analyze in-DRAM TRR implementations. U-TRR is based on the new observation that data retention failures in DRAM enable a side channel that leaks information on how TRR refreshes potential victim rows. U-TRR allows us to (i) understand how logical DRAM rows are laid out physically in silicon; (ii) study undocumented on-die TRR mechanisms; and (iii) combine (i) and (ii) to evaluate the RowHammer security guarantees of modern DRAM chips. We show how U-TRR allows us to craft RowHammer access patterns that successfully circumvent the TRR mechanisms employed in 45 DRAM modules of the three major DRAM vendors. We find that the DRAM modules we analyze are vulnerable to RowHammer, having bit flips in up to 99.9% of all DRAM rows. We make U-TRR source code openly and freely available at [106].

Lois Orosa, Abdullah Giray Yağlıkçı, Haocong Luo et al.

RowHammer is a circuit-level DRAM vulnerability where repeatedly accessing (i.e., hammering) a DRAM row can cause bit flips in physically nearby rows. The RowHammer vulnerability worsens as DRAM cell size and cell-to-cell spacing shrink. Recent studies demonstrate that modern DRAM chips, including chips previously marketed as RowHammer-safe, are even more vulnerable to RowHammer than older chips such that the required hammer count to cause a bit flip has reduced by more than 10X in the last decade. Therefore, it is essential to develop a better understanding and in-depth insights into the RowHammer vulnerability of modern DRAM chips to more effectively secure current and future systems. Our goal in this paper is to provide insights into fundamental properties of the RowHammer vulnerability that are not yet rigorously studied by prior works, but can potentially be $i$) exploited to develop more effective RowHammer attacks or $ii$) leveraged to design more effective and efficient defense mechanisms. To this end, we present an experimental characterization using 248~DDR4 and 24~DDR3 modern DRAM chips from four major DRAM manufacturers demonstrating how the RowHammer effects vary with three fundamental properties: 1)~DRAM chip temperature, 2)~aggressor row active time, and 3)~victim DRAM cell's physical location. Among our 16 new observations, we highlight that a RowHammer bit flip 1)~is very likely to occur in a bounded range, specific to each DRAM cell (e.g., 5.4% of the vulnerable DRAM cells exhibit errors in the range 70C to 90C), 2)~is more likely to occur if the aggressor row is active for longer time (e.g., RowHammer vulnerability increases by 36% if we keep a DRAM row active for 15 column accesses), and 3)~is more likely to occur in certain physical regions of the DRAM module under attack (e.g., 5% of the rows are 2x more vulnerable than the remaining 95% of the rows).

Amirali Boroumand, Saugata Ghose, Berkin Akin et al.

Emerging edge computing platforms often contain machine learning (ML) accelerators that can accelerate inference for a wide range of neural network (NN) models. These models are designed to fit within the limited area and energy constraints of the edge computing platforms, each targeting various applications (e.g., face detection, speech recognition, translation, image captioning, video analytics). To understand how edge ML accelerators perform, we characterize the performance of a commercial Google Edge TPU, using 24 Google edge NN models (which span a wide range of NN model types) and analyzing each NN layer within each model. We find that the Edge TPU suffers from three major shortcomings: (1) it operates significantly below peak computational throughput, (2) it operates significantly below its theoretical energy efficiency, and (3) its memory system is a large energy and performance bottleneck. Our characterization reveals that the one-size-fits-all, monolithic design of the Edge TPU ignores the high degree of heterogeneity both across different NN models and across different NN layers within the same NN model, leading to the shortcomings we observe. We propose a new acceleration framework called Mensa. Mensa incorporates multiple heterogeneous edge ML accelerators (including both on-chip and near-data accelerators), each of which caters to the characteristics of a particular subset of NN models and layers. During NN inference, for each NN layer, Mensa decides which accelerator to schedule the layer on, taking into account both the optimality of each accelerator for the layer and layer-to-layer communication costs. Averaged across all 24 Google edge NN models, Mensa improves energy efficiency and throughput by 3.0x and 3.1x over the Edge TPU, and by 2.4x and 4.3x over Eyeriss~v2, a state-of-the-art accelerator.

Sherif A. S. Mohamed, Mohammad-Hashem Haghbayan, Antonio Miele et al.

We propose an energy-efficient controller to minimize the energy consumption of a mobile robot by dynamically manipulating the mechanical and computational actuators of the robot. The mobile robot performs real-time vision-based applications based on an event-based camera. The actuators of the controller are CPU voltage/frequency for the computation part and motor voltage for the mechanical part. We show that independently considering speed control of the robot and voltage/frequency control of the CPU does not necessarily result in an energy-efficient solution. In fact, to obtain the highest efficiency, the computation and mechanical parts should be controlled together in synergy. We propose a fast hill-climbing optimization algorithm to allow the controller to find the best CPU/motor configuration at run-time and whenever the mobile robot is facing a new environment during its travel. Experimental results on a robot with Brushless DC Motors, Jetson TX2 board as the computing unit, and a DAVIS-346 event-based camera show that the proposed control algorithm can save battery energy by an average of 50.5%, 41%, and 30%, in low-complexity, medium-complexity, and high-complexity environments, over baselines.

Abdullah Giray Yağlıkçı, Jeremie S. Kim, Fabrice Devaux et al.

The purpose of this document is to study the security properties of the Silver Bullet algorithm against worst-case RowHammer attacks. We mathematically demonstrate that Silver Bullet, when properly configured and implemented in a DRAM chip, can securely prevent RowHammer attacks. The demonstration focuses on the most representative implementation of Silver Bullet, the patent claiming many implementation possibilities not covered in this demonstration. Our study concludes that Silver Bullet is a promising RowHammer prevention mechanism that can be configured to operate securely against RowHammer attacks at various efficiency-area tradeoff points, supporting relatively small hammer count values (e.g., 1000) and Silver Bullet table sizes (e.g., 1.06KB).

Lois Orosa, Yaohua Wang, Mohammad Sadrosadati et al.

DRAM is the dominant main memory technology used in modern computing systems. Computing systems implement a memory controller that interfaces with DRAM via DRAM commands. DRAM executes the given commands using internal components (e.g., access transistors, sense amplifiers) that are orchestrated by DRAM internal timings, which are fixed foreach DRAM command. Unfortunately, the use of fixed internal timings limits the types of operations that DRAM can perform and hinders the implementation of new functionalities and custom mechanisms that improve DRAM reliability, performance and energy. To overcome these limitations, we propose enabling programmable DRAM internal timings for controlling in-DRAM components. To this end, we design CODIC, a new low-cost DRAM substrate that enables fine-grained control over four previously fixed internal DRAM timings that are key to many DRAM operations. We implement CODIC with only minimal changes to the DRAM chip and the DDRx interface. To demonstrate the potential of CODIC, we propose two new CODIC-based security mechanisms that outperform state-of-the-art mechanisms in several ways: (1) a new DRAM Physical Unclonable Function (PUF) that is more robust and has significantly higher throughput than state-of-the-art DRAM PUFs, and (2) the first cold boot attack prevention mechanism that does not introduce any performance or energy overheads at runtime.

Nour Almadhoun Alserr, Gulce Kale, Onur Mutlu et al.

Motivation: Researchers need a rich trove of genomic datasets that they can leverage to gain a better understanding of the genetic basis of the human genome and identify associations between phenotypes and specific parts of DNA. However, sharing genomic datasets that include sensitive genetic or medical information of individuals can lead to serious privacy-related consequences if data lands in the wrong hands. Restricting access to genomic datasets is one solution, but this greatly reduces their usefulness for research purposes. To allow sharing of genomic datasets while addressing these privacy concerns, several studies propose privacy-preserving mechanisms for data sharing. Differential privacy (DP) is one of such mechanisms that formalize rigorous mathematical foundations to provide privacy guarantees while sharing aggregated statistical information about a dataset. However, it has been shown that the original privacy guarantees of DP-based solutions degrade when there are dependent tuples in the dataset, which is a common scenario for genomic datasets (due to the existence of family members). Results: In this work, we introduce a near-optimal mechanism to mitigate the vulnerabilities of the inference attacks on differentially private query results from genomic datasets including dependent tuples. We propose a utility-maximizing and privacy-preserving approach for sharing statistics by hiding selective SNPs of the family members as they participate in a genomic dataset. By evaluating our mechanism on a real-world genomic dataset, we empirically demonstrate that our proposed mechanism can achieve up to 40% better privacy than state-of-the-art DP-based solutions, while near-optimally minimizing the utility loss.

Jawad Haj-Yahya, Jeremie S. Kim, A. Giray Yaglikci et al.

To operate efficiently across a wide range of workloads with varying power requirements, a modern processor applies different current management mechanisms, which briefly throttle instruction execution while they adjust voltage and frequency to accommodate for power-hungry instructions (PHIs) in the instruction stream. Doing so 1) reduces the power consumption of non-PHI instructions in typical workloads and 2) optimizes system voltage regulators' cost and area for the common use case while limiting current consumption when executing PHIs. However, these mechanisms may compromise a system's confidentiality guarantees. In particular, we observe that multilevel side-effects of throttling mechanisms, due to PHI-related current management mechanisms, can be detected by two different software contexts (i.e., sender and receiver) running on 1) the same hardware thread, 2) co-located Simultaneous Multi-Threading (SMT) threads, and 3) different physical cores. Based on these new observations on current management mechanisms, we develop a new set of covert channels, IChannels, and demonstrate them in real modern Intel processors (which span more than 70% of the entire client and server processor market). Our analysis shows that IChannels provides more than 24x the channel capacity of state-of-the-art power management covert channels. We propose practical and effective mitigations to each covert channel in IChannels by leveraging the insights we gain through a rigorous characterization of real systems.

Ataberk Olgun, Minesh Patel, A. Giray Yağlıkçı et al.

True random number generators (TRNG) sample random physical processes to create large amounts of random numbers for various use cases, including security-critical cryptographic primitives, scientific simulations, machine learning applications, and even recreational entertainment. Unfortunately, not every computing system is equipped with dedicated TRNG hardware, limiting the application space and security guarantees for such systems. To open the application space and enable security guarantees for the overwhelming majority of computing systems that do not necessarily have dedicated TRNG hardware, we develop QUAC-TRNG. QUAC-TRNG exploits the new observation that a carefully-engineered sequence of DRAM commands activates four consecutive DRAM rows in rapid succession. This QUadruple ACtivation (QUAC) causes the bitline sense amplifiers to non-deterministically converge to random values when we activate four rows that store conflicting data because the net deviation in bitline voltage fails to meet reliable sensing margins. We experimentally demonstrate that QUAC reliably generates random values across 136 commodity DDR4 DRAM chips from one major DRAM manufacturer. We describe how to develop an effective TRNG (QUAC-TRNG) based on QUAC. We evaluate the quality of our TRNG using NIST STS and find that QUAC-TRNG successfully passes each test. Our experimental evaluations show that QUAC-TRNG generates true random numbers with a throughput of 3.44 Gb/s (per DRAM channel), outperforming the state-of-the-art DRAM-based TRNG by 15.08x and 1.41x for basic and throughput-optimized versions, respectively. We show that QUAC-TRNG utilizes DRAM bandwidth better than the state-of-the-art, achieving up to 2.03x the throughput of a throughput-optimized baseline when scaling bus frequencies to 12 GT/s.

Maciej Besta, Zur Vonarburg-Shmaria, Yannick Schaffner et al.

We propose GraphMineSuite (GMS): the first benchmarking suite for graph mining that facilitates evaluating and constructing high-performance graph mining algorithms. First, GMS comes with a benchmark specification based on extensive literature review, prescribing representative problems, algorithms, and datasets. Second, GMS offers a carefully designed software platform for seamless testing of different fine-grained elements of graph mining algorithms, such as graph representations or algorithm subroutines. The platform includes parallel implementations of more than 40 considered baselines, and it facilitates developing complex and fast mining algorithms. High modularity is possible by harnessing set algebra operations such as set intersection and difference, which enables breaking complex graph mining algorithms into simple building blocks that can be separately experimented with. GMS is supported with a broad concurrency analysis for portability in performance insights, and a novel performance metric to assess the throughput of graph mining algorithms, enabling more insightful evaluation. As use cases, we harness GMS to rapidly redesign and accelerate state-of-the-art baselines of core graph mining problems: degeneracy reordering (by up to >2x), maximal clique listing (by up to >9x), k-clique listing (by 1.1x), and subgraph isomorphism (by up to 2.5x), also obtaining better theoretical performance bounds.

Amirali Boroumand, Saugata Ghose, Berkin Akin et al.

As the need for edge computing grows, many modern consumer devices now contain edge machine learning (ML) accelerators that can compute a wide range of neural network (NN) models while still fitting within tight resource constraints. We analyze a commercial Edge TPU using 24 Google edge NN models (including CNNs, LSTMs, transducers, and RCNNs), and find that the accelerator suffers from three shortcomings, in terms of computational throughput, energy efficiency, and memory access handling. We comprehensively study the characteristics of each NN layer in all of the Google edge models, and find that these shortcomings arise from the one-size-fits-all approach of the accelerator, as there is a high amount of heterogeneity in key layer characteristics both across different models and across different layers in the same model. We propose a new acceleration framework called Mensa. Mensa incorporates multiple heterogeneous ML edge accelerators (including both on-chip and near-data accelerators), each of which caters to the characteristics of a particular subset of models. At runtime, Mensa schedules each layer to run on the best-suited accelerator, accounting for both efficiency and inter-layer dependencies. As we analyze the Google edge NN models, we discover that all of the layers naturally group into a small number of clusters, which allows us to design an efficient implementation of Mensa for these models with only three specialized accelerators. Averaged across all 24 Google edge models, Mensa improves energy efficiency and throughput by 3.0x and 3.1x over the Edge TPU, and by 2.4x and 4.3x over Eyeriss v2, a state-of-the-art accelerator.

Abdullah Giray Yağlıkçı, Minesh Patel, Jeremie S. Kim et al.

Aggressive memory density scaling causes modern DRAM devices to suffer from RowHammer, a phenomenon where rapidly activating a DRAM row can cause bit-flips in physically-nearby rows. Recent studies demonstrate that modern DRAM chips, including chips previously marketed as RowHammer-safe, are even more vulnerable to RowHammer than older chips. Many works show that attackers can exploit RowHammer bit-flips to reliably mount system-level attacks to escalate privilege and leak private data. Therefore, it is critical to ensure RowHammer-safe operation on all DRAM-based systems. Unfortunately, state-of-the-art RowHammer mitigation mechanisms face two major challenges. First, they incur increasingly higher performance and/or area overheads when applied to more vulnerable DRAM chips. Second, they require either proprietary information about or modifications to the DRAM chip design. In this paper, we show that it is possible to efficiently and scalably prevent RowHammer bit-flips without knowledge of or modification to DRAM internals. We introduce BlockHammer, a low-cost, effective, and easy-to-adopt RowHammer mitigation mechanism that overcomes the two key challenges by selectively throttling memory accesses that could otherwise cause RowHammer bit-flips. The key idea of BlockHammer is to (1) track row activation rates using area-efficient Bloom filters and (2) use the tracking data to ensure that no row is ever activated rapidly enough to induce RowHammer bit-flips. By doing so, BlockHammer (1) makes it impossible for a RowHammer bit-flip to occur and (2) greatly reduces a RowHammer attack's impact on the performance of co-running benign applications. Compared to state-of-the-art RowHammer mitigation mechanisms, BlockHammer provides competitive performance and energy when the system is not under a RowHammer attack and significantly better performance and energy when the system is under attack.

Mohammed Alser, Jeremie S. Kim, Nour Almadhoun Alserr et al.

Background: Early detection and isolation of COVID-19 patients are essential for successful implementation of mitigation strategies and eventually curbing the disease spread. With a limited number of daily COVID-19 tests performed in every country, simulating the COVID-19 spread along with the potential effect of each mitigation strategy currently remains one of the most effective ways in managing the healthcare system and guiding policy-makers. Methods: We introduce COVIDHunter, a flexible and accurate COVID-19 outbreak simulation model that evaluates the current mitigation measures that are applied to a region and provides suggestions on what strength the upcoming mitigation measure should be. The key idea of COVIDHunter is to quantify the spread of COVID-19 in a geographical region by simulating the average number of new infections caused by an infected person considering the effect of external factors, such as environmental conditions (e.g., climate, temperature, humidity) and mitigation measures. Results: Using Switzerland as a case study, COVIDHunter estimates that if the policy-makers relax the mitigation measures by 50% for 30 days then both the daily capacity need for hospital beds and daily number of deaths increase exponentially by an average of 5.1x, who may occupy ICU beds and ventilators for a period of time. Unlike existing models, the COVIDHunter model accurately monitors and predicts the daily number of cases, hospitalizations, and deaths due to COVID-19. Our model is flexible to configure and simple to modify for modeling different scenarios under different environmental conditions and mitigation measures. Availability: We release the source code of the COVIDHunter implementation at https://github.com/CMU- SAFARI/COVIDHunter and show how to flexibly configure our model for any scenario and easily extend it for different measures and conditions than we account for.

Muhammad Shafique, Mahum Naseer, Theocharis Theocharides et al.

Machine Learning (ML) techniques have been rapidly adopted by smart Cyber-Physical Systems (CPS) and Internet-of-Things (IoT) due to their powerful decision-making capabilities. However, they are vulnerable to various security and reliability threats, at both hardware and software levels, that compromise their accuracy. These threats get aggravated in emerging edge ML devices that have stringent constraints in terms of resources (e.g., compute, memory, power/energy), and that therefore cannot employ costly security and reliability measures. Security, reliability, and vulnerability mitigation techniques span from network security measures to hardware protection, with an increased interest towards formal verification of trained ML models. This paper summarizes the prominent vulnerabilities of modern ML systems, highlights successful defenses and mitigation techniques against these vulnerabilities, both at the cloud (i.e., during the ML training phase) and edge (i.e., during the ML inference stage), discusses the implications of a resource-constrained design on the reliability and security of the system, identifies verification methodologies to ensure correct system behavior, and describes open research challenges for building secure and reliable ML systems at both the edge and the cloud.

Jeremie S. Kim, Minesh Patel, A. Giray Yaglikci et al.

In order to shed more light on how RowHammer affects modern and future devices at the circuit-level, we first present an experimental characterization of RowHammer on 1580 DRAM chips (408x DDR3, 652x DDR4, and 520x LPDDR4) from 300 DRAM modules (60x DDR3, 110x DDR4, and 130x LPDDR4) with RowHammer protection mechanisms disabled, spanning multiple different technology nodes from across each of the three major DRAM manufacturers. Our studies definitively show that newer DRAM chips are more vulnerable to RowHammer: as device feature size reduces, the number of activations needed to induce a RowHammer bit flip also reduces, to as few as 9.6k (4.8k to two rows each) in the most vulnerable chip we tested. We evaluate five state-of-the-art RowHammer mitigation mechanisms using cycle-accurate simulation in the context of real data taken from our chips to study how the mitigation mechanisms scale with chip vulnerability. We find that existing mechanisms either are not scalable or suffer from prohibitively large performance overheads in projected future devices given our observed trends of RowHammer vulnerability. Thus, it is critical to research more effective solutions to RowHammer.

Behzad Salami, Erhan Baturay Onural, Ismail Emir Yuksel et al.

We empirically evaluate an undervolting technique, i.e., underscaling the circuit supply voltage below the nominal level, to improve the power-efficiency of Convolutional Neural Network (CNN) accelerators mapped to Field Programmable Gate Arrays (FPGAs). Undervolting below a safe voltage level can lead to timing faults due to excessive circuit latency increase. We evaluate the reliability-power trade-off for such accelerators. Specifically, we experimentally study the reduced-voltage operation of multiple components of real FPGAs, characterize the corresponding reliability behavior of CNN accelerators, propose techniques to minimize the drawbacks of reduced-voltage operation, and combine undervolting with architectural CNN optimization techniques, i.e., quantization and pruning. We investigate the effect of environmental temperature on the reliability-power trade-off of such accelerators. We perform experiments on three identical samples of modern Xilinx ZCU102 FPGA platforms with five state-of-the-art image classification CNN benchmarks. This approach allows us to study the effects of our undervolting technique for both software and hardware variability. We achieve more than 3X power-efficiency (GOPs/W) gain via undervolting. 2.6X of this gain is the result of eliminating the voltage guardband region, i.e., the safe voltage region below the nominal level that is set by FPGA vendor to ensure correct functionality in worst-case environmental and circuit conditions. 43% of the power-efficiency gain is due to further undervolting below the guardband, which comes at the cost of accuracy loss in the CNN accelerator. We evaluate an effective frequency underscaling technique that prevents this accuracy loss, and find that it reduces the power-efficiency gain from 43% to 25%.