Comp. Arch. - Lecture 28: Systolic Array Architectures (Fall 2025)

Item: Comp. Arch. - Lecture 28: Systolic Array Architectures (Fall 2025)
Rating: 5
Author: PostSingularity Institute

by Onur Mutlu

Formal & Physical Sciences Computing & Cybersecurity UYFComputer architecture and logic design UYFPParallel processing

🎙 Onur Mutlu 👥 64K 📅 January 8, 2026 ⏱ 42 min 👁 4K 📄 lecture 🧭 2026-08-15

Available in: English (current) Français

Keywords

systolic arrayexecution modelconvolutionmatrix multiplicationaccelerator

Summary

This lecture, part of the Computer Architecture course at ETH Zürich, provides a comprehensive introduction to systolic array architectures. Professor Onur Mutlu begins by contrasting general-purpose processors with special-purpose accelerators, positioning systolic arrays as a specialized execution model. He explains the motivation behind systolic arrays: simple and regular design, high concurrency, and balanced computation and I/O. The core idea is to replace a single processing element with a regular array of processing elements, where data flows rhythmically from memory through the array, similar to blood flow in the body. This model minimizes memory accesses by maximizing computation on each data element. The lecture details the differences from pipelining and other execution models like Von Neumann and dataflow. A key example is convolution, heavily used in image processing and modern machine learning. Mutlu illustrates how convolutional neural networks (CNNs) rely on convolution operations, which can be efficiently implemented as matrix multiplications on systolic arrays. He references the historical impact of GPUs in enabling deep learning, citing the 2012 AlexNet breakthrough. The lecture concludes by emphasizing the fundamental nature of systolic arrays and their relevance in contemporary machine learning accelerators.

188 words

Critical Evaluation

Value of the Information & Strength of the Argument

The lecture provides high-value information by clearly explaining the principles, motivations, and applications of systolic arrays. The argumentation is solid, building from basic concepts to complex applications. Mutlu effectively uses analogies (e.g., blood flow) and concrete examples (convolution, CNNs) to illustrate abstract ideas. He also connects the material to broader trends in computer architecture and machine learning, enhancing its relevance. The reasoning is logical and well-structured, making a compelling case for the importance of systolic arrays.

Scientific Rigor, Source Quality, Title Accuracy

The lecture demonstrates high scientific rigor. Mutlu references seminal papers, including Kung’s work on systolic arrays, and provides recommended readings from reputable sources. The course materials and slides are available online, and the lecture is part of a well-established university course. The title accurately reflects the content, and the lecture stays focused on the topic. No comments were provided for analysis.

152 words

Title / Content Match

The title accurately reflects the content, which focuses on systolic array architectures.

Quality & Reliability

9/10

Lecture by a renowned professor in computer architecture, based on established concepts and peer-reviewed research, with references to seminal papers and course materials.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

A Modern Primer on Processing in Memory — Recommended reading for understanding memory-centric computing, related to systolic arrays' goal of balancing computation and memory.
Memory-Centric Computing: Solving Computing's Memory Problem — Recommended reading on memory-centric computing, relevant to the motivation for systolic arrays.
Memory-Centric Computing: Recent Advances in Processing-in-DRAM — Recommended reading on processing-in-DRAM, related to memory-centric computing.
Intelligent Architectures for Intelligent Computing Systems — Recommended reading on intelligent architectures, relevant to the lecture's themes.
RowHammer: A Retrospective — Recommended reading on RowHammer, a memory reliability issue, related to memory-centric computing.
Fundamentally Understanding and Solving RowHammer — Recommended reading on RowHammer, relevant to memory reliability.
Accelerating Genome Analysis via Algorithm-Architecture Co-Design — Recommended reading on algorithm-architecture co-design, showing application of specialized architectures.
From Molecules to Genomic Variations: Accelerating Genome Analysis via Intelligent Algorithms and Architectures — Recommended reading on genome analysis acceleration, related to specialized architectures.
Course Schedule — Course page for Computer Architecture, Fall 2025.
Lecture Slides (PDF) — Slides for this lecture.
Lecture Slides (PPTX) — Slides for this lecture in PowerPoint format.

Concurring Sources

Systolic array - Wikipedia — Provides general information on systolic arrays, consistent with the lecture's content.
Google TPU paper — Shows a real-world implementation of systolic arrays in machine learning accelerators, supporting the lecture's claims.

External References

Contribution & Novelties

This lecture provides a clear and comprehensive overview of systolic array architectures, emphasizing their role as a fundamental execution model. It connects historical motivations to modern machine learning accelerators, illustrating the evolution of the concept. The lecture’s strength lies in its pedagogical approach, using analogies and concrete examples to explain complex ideas.

Pour aller plus loin :

Systolic array - Wikipedia — Overview of systolic arrays and their applications.
Convolutional neural network - Wikipedia — Background on CNNs, which heavily use convolution operations.
H. T. Kung’s paper on systolic arrays — Original paper by H. T. Kung, one of the inventors of systolic arrays.
Google TPU paper — Describes the use of systolic arrays in Google’s Tensor Processing Unit.

119 words

Radar Profile

The radar profile shows high scores across all dimensions, indicating a well-rounded and reliable lecture. The high scores in information quantity and quality reflect the depth and accuracy of the content, while the technical level is appropriate for an advanced computer architecture course. The overall reliability is excellent, making this a valuable educational resource.

Reliability 9/10