arXiv subjects

Geraldo F. Oliveira

Publications and source records attributed to Geraldo F. Oliveira.

At least 37 records · Page 2Linked to original sources

arXiv2024

Sectored DRAM: A Practical Energy-Efficient and High-Performance Fine-Grained DRAM Architecture

We propose Sectored DRAM, a new, low-overhead DRAM substrate that reduces wasted energy by enabling fine-grained DRAM data transfers and DRAM row activation. Sectored DRAM leverages two key ideas to enable fine-grained data transfers and row activation at low chip area cost. First, a cache block transfer between main memory and the memory controller happens in a fixed number of clock cycles where only a small portion of the cache block (a word) is transferred in each cycle. Sectored DRAM augments the memory controller and the DRAM chip to execute cache block transfers in a variable number of clock cycles based on the workload access pattern with minor modifications to the memory controller's and the DRAM chip's circuitry. Second, a large DRAM row, by design, is already partitioned into smaller independent physically isolated regions. Sectored DRAM provides the memory controller with the ability to activate each such region based on the workload access pattern via small modifications to the DRAM chip's array access circuitry. Activating smaller regions of a large row relaxes DRAM power delivery constraints and allows the memory controller to schedule DRAM accesses faster. Compared to a system with coarse-grained DRAM, Sectored DRAM reduces the DRAM energy consumption of highly-memory-intensive workloads by up to (on average) 33% (20%) while improving their performance by up to (on average) 36% (17%). Sectored DRAM's DRAM energy savings, combined with its system performance improvement, allows system-wide energy savings of up to 23%. Sectored DRAM's DRAM chip area overhead is 1.7% the area of a modern DDR4 chip. We hope and believe that Sectored DRAM's ideas and results will help to enable more efficient and high-performance memory systems. To this end, we open source Sectored DRAM at https://github.com/CMU-SAFARI/Sectored-DRAM.

cs.AR↗

arXiv2024

Simultaneous Many-Row Activation in Off-the-Shelf DRAM Chips: Experimental Characterization and Analysis

We experimentally analyze the computational capability of commercial off-the-shelf (COTS) DRAM chips and the robustness of these capabilities under various timing delays between DRAM commands, data patterns, temperature, and voltage levels. We extensively characterize 120 COTS DDR4 chips from two major manufacturers. We highlight four key results of our study. First, COTS DRAM chips are capable of 1) simultaneously activating up to 32 rows (i.e., simultaneous many-row activation), 2) executing a majority of X (MAJX) operation where X>3 (i.e., MAJ5, MAJ7, and MAJ9 operations), and 3) copying a DRAM row (concurrently) to up to 31 other DRAM rows, which we call Multi-RowCopy. Second, storing multiple copies of MAJX's input operands on all simultaneously activated rows drastically increases the success rate (i.e., the percentage of DRAM cells that correctly perform the computation) of the MAJX operation. For example, MAJ3 with 32-row activation (i.e., replicating each MAJ3's input operands 10 times) has a 30.81% higher average success rate than MAJ3 with 4-row activation (i.e., no replication). Third, data pattern affects the success rate of MAJX and Multi-RowCopy operations by 11.52% and 0.07% on average. Fourth, simultaneous many-row activation, MAJX, and Multi-RowCopy operations are highly resilient to temperature and voltage changes, with small success rate variations of at most 2.13% among all tested operations. We believe these empirical results demonstrate the promising potential of using DRAM as a computation substrate. To aid future research and development, we open-source our infrastructure at https://github.com/CMU-SAFARI/SiMRA-DRAM.

cs.AR↗

arXiv2024

ABACuS: All-Bank Activation Counters for Scalable and Low Overhead RowHammer Mitigation

We introduce ABACuS, a new low-cost hardware-counter-based RowHammer mitigation technique that performance-, energy-, and area-efficiently scales with worsening RowHammer vulnerability. We observe that both benign workloads and RowHammer attacks tend to access DRAM rows with the same row address in multiple DRAM banks at around the same time. Based on this observation, ABACuS's key idea is to use a single shared row activation counter to track activations to the rows with the same row address in all DRAM banks. Unlike state-of-the-art RowHammer mitigation mechanisms that implement a separate row activation counter for each DRAM bank, ABACuS implements fewer counters (e.g., only one) to track an equal number of aggressor rows. Our evaluations show that ABACuS securely prevents RowHammer bitflips at low performance/energy overhead and low area cost. We compare ABACuS to four state-of-the-art mitigation mechanisms. At a near-future RowHammer threshold of 1000, ABACuS incurs only 0.58% (0.77%) performance and 1.66% (2.12%) DRAM energy overheads, averaged across 62 single-core (8-core) workloads, requiring only 9.47 KiB of storage per DRAM rank. At the RowHammer threshold of 1000, the best prior low-area-cost mitigation mechanism incurs 1.80% higher average performance overhead than ABACuS, while ABACuS requires 2.50X smaller chip area to implement. At a future RowHammer threshold of 125, ABACuS performs very similarly to (within 0.38% of the performance of) the best prior performance- and energy-efficient RowHammer mitigation mechanism while requiring 22.72X smaller chip area. ABACuS is freely and openly available at https://github.com/CMU-SAFARI/ABACuS.

cs.CR↗

arXiv2024

Functionally-Complete Boolean Logic in Real DRAM Chips: Experimental Characterization and Analysis

Processing-using-DRAM (PuD) is an emerging paradigm that leverages the analog operational properties of DRAM circuitry to enable massively parallel in-DRAM computation. PuD has the potential to reduce or eliminate costly data movement between processing elements and main memory. Prior works experimentally demonstrate three-input MAJ (MAJ3) and two-input AND and OR operations in commercial off-the-shelf (COTS) DRAM chips. Yet, demonstrations on COTS DRAM chips do not provide a functionally complete set of operations. We experimentally demonstrate that COTS DRAM chips are capable of performing 1) functionally-complete Boolean operations: NOT, NAND, and NOR and 2) many-input (i.e., more than two-input) AND and OR operations. We present an extensive characterization of new bulk bitwise operations in 256 off-the-shelf modern DDR4 DRAM chips. We evaluate the reliability of these operations using a metric called success rate: the fraction of correctly performed bitwise operations. Among our 19 new observations, we highlight four major results. First, we can perform the NOT operation on COTS DRAM chips with a 98.37% success rate on average. Second, we can perform up to 16-input NAND, NOR, AND, and OR operations on COTS DRAM chips with high reliability (e.g., 16-input NAND, NOR, AND, and OR with an average success rate of 94.94%, 95.87%, 94.94%, and 95.85%, respectively). Third, data pattern only slightly affects bitwise operations. Our results show that executing NAND, NOR, AND, and OR operations with random data patterns decreases the success rate compared to all logic-1/logic-0 patterns by 1.39%, 1.97%, 1.43%, and 1.98%, respectively. Fourth, bitwise operations are highly resilient to temperature changes, with small success rate fluctuations of at most 1.66% when the temperature is increased from 50C to 95C. We open-source our infrastructure at https://github.com/CMU-SAFARI/FCDRAM

cs.AR↗

arXiv2024

PULSAR: Simultaneous Many-Row Activation for Reliable and High-Performance Computing in Off-the-Shelf DRAM Chips

Data movement between the processor and the main memory is a first-order obstacle against improving performance and energy efficiency in modern systems. To address this obstacle, Processing-using-Memory (PuM) is a promising approach where bulk-bitwise operations are performed leveraging intrinsic analog properties within the DRAM array and massive parallelism across DRAM columns. Unfortunately, 1) modern off-the-shelf DRAM chips do not officially support PuM operations, and 2) existing techniques of performing PuM operations on off-the-shelf DRAM chips suffer from two key limitations. First, these techniques have low success rates, i.e., only a small fraction of DRAM columns can correctly execute PuM operations because they operate beyond manufacturer-recommended timing constraints, causing these operations to be highly susceptible to noise and process variation. Second, these techniques have limited compute primitives, preventing them from fully leveraging parallelism across DRAM columns and thus hindering their performance benefits. We propose PULSAR, a new technique to enable high-success-rate and high-performance PuM operations in off-the-shelf DRAM chips. PULSAR leverages our new observation that a carefully crafted sequence of DRAM commands simultaneously activates up to 32 DRAM rows. PULSAR overcomes the limitations of existing techniques by 1) replicating the input data to improve the success rate and 2) enabling new bulk bitwise operations (e.g., many-input majority, Multi-RowInit, and Bulk-Write) to improve the performance. Our analysis on 120 off-the-shelf DDR4 chips from two major manufacturers shows that PULSAR achieves a 24.18% higher success rate and 121% higher performance over seven arithmetic-logic operations compared to FracDRAM, a state-of-the-art off-the-shelf DRAM-based PuM technique.

cs.AR↗

arXiv2024

PUMA: Efficient and Low-Cost Memory Allocation and Alignment Support for Processing-Using-Memory Architectures

Processing-using-DRAM (PUD) architectures impose a restrictive data layout and alignment for their operands, where source and destination operands (i) must reside in the same DRAM subarray (i.e., a group of DRAM rows sharing the same row buffer and row decoder) and (ii) are aligned to the boundaries of a DRAM row. However, standard memory allocation routines (i.e., malloc, posix_memalign, and huge pages-based memory allocation) fail to meet the data layout and alignment requirements for PUD architectures to operate successfully. To allow the memory allocation API to influence the OS memory allocator and ensure that memory objects are placed within specific DRAM subarrays, we propose a new lazy data allocation routine (in the kernel) for PUD memory objects called PUMA. The key idea of PUMA is to use the internal DRAM mapping information together with huge pages and then split huge pages into finer-grained allocation units that are (i) aligned to the page address and size and (ii) virtually contiguous. We implement PUMA as a kernel module using QEMU and emulate a RISC-V machine running Fedora 33 with v5.9.0 Linux Kernel. We emulate the implementation of a PUD system capable of executing row copy operations (as in RowClone) and Boolean AND/OR/NOT operations (as in Ambit). In our experiments, such an operation is performed in the host CPU if a given operation cannot be executed in our PUD substrate (due to data misalignment). PUMA significantly outperforms the baseline memory allocators for all evaluated microbenchmarks and allocation sizes.

cs.AR↗

arXiv2024

MIMDRAM: An End-to-End Processing-Using-DRAM System for High-Throughput, Energy-Efficient and Programmer-Transparent Multiple-Instruction Multiple-Data Processing

Processing-using-DRAM (PUD) is a processing-in-memory (PIM) approach that uses a DRAM array's massive internal parallelism to execute very-wide data-parallel operations, in a single-instruction multiple-data (SIMD) fashion. However, DRAM rows' large and rigid granularity limit the effectiveness and applicability of PUD in three ways. First, since applications have varying degrees of SIMD parallelism, PUD execution often leads to underutilization, throughput loss, and energy waste. Second, most PUD architectures are limited to the execution of parallel map operations. Third, the need to feed the wide DRAM row with tens of thousands of data elements combined with the lack of adequate compiler support for PUD systems create a programmability barrier. Our goal is to design a flexible PUD system that overcomes the limitations caused by the large and rigid granularity of PUD. To this end, we propose MIMDRAM, a hardware/software co-designed PUD system that introduces new mechanisms to allocate and control only the necessary resources for a given PUD operation. The key idea of MIMDRAM is to leverage fine-grained DRAM (i.e., the ability to independently access smaller segments of a large DRAM row) for PUD computation. MIMDRAM exploits this key idea to enable a multiple-instruction multiple-data (MIMD) execution model in each DRAM subarray. We evaluate MIMDRAM using twelve real-world applications and 495 multi-programmed application mixes. Our evaluation shows that MIMDRAM provides 34x the performance, 14.3x the energy efficiency, 1.7x the throughput, and 1.3x the fairness of a state-of-the-art PUD framework, along with 30.6x and 6.8x the energy efficiency of a high-end CPU and GPU, respectively. MIMDRAM adds small area cost to a DRAM chip (1.11%) and CPU die (0.6%).

cs.AR↗

arXiv2024

Spatial Variation-Aware Read Disturbance Defenses: Experimental Analysis of Real DRAM Chips and Implications on Future Solutions

Read disturbance in modern DRAM chips is a widespread phenomenon and is reliably used for breaking memory isolation, a fundamental building block for building robust systems. RowHammer and RowPress are two examples of read disturbance in DRAM where repeatedly accessing (hammering) or keeping active (pressing) a memory location induces bitflips in other memory locations. Unfortunately, shrinking technology node size exacerbates read disturbance in DRAM chips over generations. As a result, existing defense mechanisms suffer from significant performance and energy overheads, limited effectiveness, or prohibitively high hardware complexity. In this paper, we tackle these shortcomings by leveraging the spatial variation in read disturbance across different memory locations in real DRAM chips. To do so, we 1) present the first rigorous real DRAM chip characterization study of spatial variation of read disturbance and 2) propose Svärd, a new mechanism that dynamically adapts the aggressiveness of existing solutions based on the row-level read disturbance profile. Our experimental characterization on 144 real DDR4 DRAM chips representing 10 chip designs demonstrates a large variation in read disturbance vulnerability across different memory locations: in the part of memory with the worst read disturbance vulnerability, 1) up to 2x the number of bitflips can occur and 2) bitflips can occur at an order of magnitude fewer accesses, compared to the memory locations with the least vulnerability to read disturbance. Svärd leverages this variation to reduce the overheads of five state-of-the-art read disturbance solutions, and thus significantly increases system performance.

cs.CR↗

arXiv2023

Extending Memory Capacity in Consumer Devices with Emerging Non-Volatile Memory: An Experimental Study

The number and diversity of consumer devices are growing rapidly, alongside their target applications' memory consumption. Unfortunately, DRAM scalability is becoming a limiting factor to the available memory capacity in consumer devices. As a potential solution, manufacturers have introduced emerging non-volatile memories (NVMs) into the market, which can be used to increase the memory capacity of consumer devices by augmenting or replacing DRAM. Since entirely replacing DRAM with NVM in consumer devices imposes large system integration and design challenges, recent works propose extending the total main memory space available to applications by using NVM as swap space for DRAM. However, no prior work analyzes the implications of enabling a real NVM-based swap space in real consumer devices. In this work, we provide the first analysis of the impact of extending the main memory space of consumer devices using off-the-shelf NVMs. We extensively examine system performance and energy consumption when the NVM device is used as swap space for DRAM main memory to effectively extend the main memory capacity. For our analyses, we equip real web-based Chromebook computers with the Intel Optane SSD, which is a state-of-the-art low-latency NVM-based SSD device. We compare the performance and energy consumption of interactive workloads running on our Chromebook with NVM-based swap space, where the Intel Optane SSD capacity is used as swap space to extend main memory capacity, against two state-of-the-art systems: (i) a baseline system with double the amount of DRAM than the system with the NVM-based swap space; and (ii) a system where the Intel Optane SSD is naively replaced with a state-of-the-art (yet slower) off-the-shelf NAND-flash-based SSD, which we use as a swap space of equivalent size as the NVM-based swap space.

cs.AR↗

arXiv2023

Casper: Accelerating Stencil Computation using Near-cache Processing

Stencil computation is one of the most used kernels in a wide variety of scientific applications, ranging from large-scale weather prediction to solving partial differential equations. Stencil computations are characterized by three unique properties: (1) low arithmetic intensity, (2) limited temporal data reuse, and (3) regular and predictable data access pattern. As a result, stencil computations are typically bandwidth-bound workloads, which only experience limited benefits from the deep cache hierarchy of modern CPUs. In this work, we propose Casper, a near-cache accelerator consisting of specialized stencil compute units connected to the last-level cache (LLC) of a traditional CPU. Casper is based on two key ideas: (1) avoiding the cost of moving rarely reused data through the cache hierarchy, and (2) exploiting the regularity of the data accesses and the inherent parallelism of the stencil computation to increase the overall performance. With minimal changes in LLC address decoding logic and data placement, Casper performs stencil computations at the peak bandwidth of the LLC. We show that, by tightly coupling lightweight stencil compute units near to LLC, Casper improves the performance of stencil kernels by 1.65x on average, while reducing the energy consumption by 35% compared to a commercial high-performance multi-core processor. Moreover, Casper provides a 37x improvement in performance-per-area compared to a state-of-the-art GPU.

cs.AR↗

arXiv2023

An Experimental Evaluation of Machine Learning Training on a Real Processing-in-Memory System

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.

cs.AR↗

arXiv2023

TransPimLib: A Library for Efficient Transcendental Functions on Processing-in-Memory Systems

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}.

cs.MS↗

arXiv2023

DAMOV: A New Methodology and Benchmark Suite for Evaluating Data Movement Bottlenecks

Data movement between the CPU and main memory is a first-order obstacle against improving performance, scalability, and energy efficiency in modern systems. Computer systems employ a range of techniques to reduce overheads tied to data movement, spanning from traditional mechanisms (e.g., deep multi-level cache hierarchies, aggressive hardware prefetchers) to emerging techniques such as Near-Data Processing (NDP), where some computation is moved close to memory. Our goal is to methodically identify potential sources of data movement over a broad set of applications and to comprehensively compare traditional compute-centric data movement mitigation techniques to more memory-centric techniques, thereby developing a rigorous understanding of the best techniques to mitigate each source of data movement. With this goal in mind, we perform the first large-scale characterization of a wide variety of applications, across a wide range of application domains, to identify fundamental program properties that lead to data movement to/from main memory. We develop the first systematic methodology to classify applications based on the sources contributing to data movement bottlenecks. From our large-scale characterization of 77K functions across 345 applications, we select 144 functions to form the first open-source benchmark suite (DAMOV) for main memory data movement studies. We select a diverse range of functions that (1) represent different types of data movement bottlenecks, and (2) come from a wide range of application domains. Using NDP as a case study, we identify new insights about the different data movement bottlenecks and use these insights to determine the most suitable data movement mitigation mechanism for a particular application. We open-source DAMOV and the complete source code for our new characterization methodology at https://github.com/CMU-SAFARI/DAMOV.

cs.AR↗

arXiv2023

Benchmarking Memory-Centric Computing Systems: Analysis of Real Processing-in-Memory Hardware

Many modern workloads such as neural network inference and graph processing are fundamentally memory-bound. For such workloads, data movement between memory and CPU cores imposes a significant overhead in terms of both latency and energy. A major reason is that this communication happens through a narrow bus with high latency and limited bandwidth, and the low data reuse in memory-bound workloads is insufficient to amortize the cost of memory access. Fundamentally addressing this data movement bottleneck requires a paradigm where the memory system assumes an active role in computing by integrating processing capabilities. This paradigm is known as processing-in-memory (PIM). Recent research explores different forms of PIM architectures, motivated by the emergence of new technologies that integrate memory with a logic layer, where processing elements can be easily placed. Past works evaluate these architectures in simulation or, at best, with simplified hardware prototypes. In contrast, the UPMEM company has designed and manufactured the first publicly-available real-world PIM architecture. The UPMEM PIM architecture combines traditional DRAM memory arrays with general-purpose in-order cores, called DRAM Processing Units (DPUs), integrated in the same chip. This paper presents key takeaways from the first comprehensive analysis of the first publicly-available real-world PIM architecture. We provide four key takeaways about the UPMEM PIM architecture, which stem from our study. More insights about suitability of different workloads to the PIM system, programming recommendations for software designers, and suggestions and hints for hardware and architecture designers of future PIM systems are available in arXiv:2105.03814

cs.AR↗

arXiv2023

Accelerating Neural Network Inference with Processing-in-DRAM: From the Edge to the Cloud

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.

cs.AR↗

arXiv2022

ALP: Alleviating CPU-Memory Data Movement Overheads in Memory-Centric Systems

Partitioning applications between NDP and host CPU cores causes inter-segment data movement overhead, which is caused by moving data generated from one segment (e.g., instructions, functions) and used in consecutive segments. Prior works take two approaches to this problem. The first class of works maps segments to NDP or host cores based on the properties of each segment, neglecting the inter-segment data movement overhead. The second class of works partitions applications based on the overall memory bandwidth saving of each segment, and does not offload each segment to the best-fitting core if they incur high inter-segment data movement. We show that 1) mapping each segment to its best-fitting core ideally can provide substantial benefits, and 2) the inter-segment data movement reduces this benefit significantly. To this end, we introduce ALP, a new programmer-transparent technique to leverage the performance benefits of NDP by alleviating the inter-segment data movement overhead between host and memory and enabling efficient partitioning of applications. ALP alleviates the inter-segment data movement overhead by proactively and accurately transferring the required data between the segments. This is based on the key observation that the instructions that generate the inter-segment data stay the same across different executions of a program on different inputs. ALP uses a compiler pass to identify these instructions and uses specialized hardware to transfer data between the host and NDP cores at runtime. ALP efficiently maps application segments to either host or NDP considering 1) the properties of each segment, 2) the inter-segment data movement overhead, and 3) whether this overhead can be alleviated in a timely manner. We evaluate ALP across a wide range of workloads and show on average 54.3% and 45.4% speedup compared to only-host CPU or only-NDP executions, respectively.

cs.AR↗

arXiv2022

Flash-Cosmos: In-Flash Bulk Bitwise Operations Using Inherent Computation Capability of NAND Flash Memory

Bulk bitwise operations, i.e., bitwise operations on large bit vectors, are prevalent in a wide range of important application domains, including databases, graph processing, genome analysis, cryptography, and hyper-dimensional computing. In conventional systems, the performance and energy efficiency of bulk bitwise operations are bottlenecked by data movement between the compute units and the memory hierarchy. In-flash processing (i.e., processing data inside NAND flash chips) has a high potential to accelerate bulk bitwise operations by fundamentally reducing data movement through the entire memory hierarchy. We identify two key limitations of the state-of-the-art in-flash processing technique for bulk bitwise operations; (i) it falls short of maximally exploiting the bit-level parallelism of bulk bitwise operations; (ii) it is unreliable because it does not consider the highly error-prone nature of NAND flash memory. We propose Flash-Cosmos (Flash Computation with One-Shot Multi-Operand Sensing), a new in-flash processing technique that significantly increases the performance and energy efficiency of bulk bitwise operations while providing high reliability. Flash-Cosmos introduces two key mechanisms that can be easily supported in modern NAND flash chips: (i) Multi-Wordline Sensing (MWS), which enables bulk bitwise operations on a large number of operands with a single sensing operation, and (ii) Enhanced SLC-mode Programming (ESP), which enables reliable computation inside NAND flash memory. We demonstrate the feasibility of performing bulk bitwise operations with high reliability in Flash-Cosmos by testing 160 real 3D NAND flash chips. Our evaluation shows that Flash-Cosmos improves average performance and energy efficiency by 3.5x/32x and 3.3x/95x, respectively, over the state-of-the-art in-flash/outside-storage processing techniques across three real-world applications.

cs.AR↗

arXiv2022

Machine Learning Training on a Real Processing-in-Memory System

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.

cs.AR↗