Seminar in Computer Architecture - Lecture 2: Memory-Centric Computing (Fall 2025)

Item: Seminar in Computer Architecture - Lecture 2: Memory-Centric Computing (Fall 2025)
Rating: 5
Author: PostSingularity Institute

by Onur Mutlu

Formal & Physical Sciences Computing & Cybersecurity UYComputer science UYFComputer architecture and logic design

🎙 Onur Mutlu 👥 64K 📅 September 26, 2025 ⏱ 110 min 👁 1K 📄 lecture 🧭 2026-08-16

Available in: English (current) Français

Keywords

memory-centric computingdata movementprocessing-in-memoryenergy efficiencycomputer architecture

Summary

In this lecture, Prof. Onur Mutlu introduces the concept of memory-centric computing as a paradigm shift to address the growing memory bottleneck in modern computing systems. He begins by highlighting the severe energy and performance costs associated with data movement, citing studies showing that over 60% of system energy is spent on data movement in mobile systems and over 90% in some machine learning accelerators. He argues that current processor-centric architectures are fundamentally inefficient, with most hardware dedicated to storing and moving data rather than computation. Mutlu proposes a more memory-centric approach, where computation is performed near or inside memory, reducing data movement and improving efficiency. He discusses various levels of the memory hierarchy, from DRAM to storage, and introduces concepts like processing-in-memory (PIM) and intelligent memory systems. The lecture also touches on the challenges of technology scaling and the need for robust system design. He emphasizes the importance of rethinking system design from a blank slate, using the analogy of a 10-year-old child to encourage fresh perspectives. The talk concludes with a call for more research in this area and mentions ongoing projects in his group, such as RowHammer mitigation and genome analysis acceleration.

195 words

Critical Evaluation

Value of the Information & Strength of the Argument

The lecture provides high-value insights into the memory bottleneck problem, supported by quantitative data from studies by Google and others. The argumentation is solid, logically progressing from the problem statement to potential solutions. Mutlu effectively uses analogies and examples to make the case for memory-centric computing, and he addresses counterarguments by acknowledging the complexity of adoption. The presentation is persuasive and well-supported by references to his own research and industry studies.

Scientific Rigor, Source Quality, Title Accuracy

The lecture demonstrates high scientific rigor, with references to peer-reviewed papers and technical reports. The sources cited are credible and directly relevant to the topic. The title accurately reflects the content, and the lecture is well-structured. The presentation is based on extensive research and includes references to specific studies, such as the Google data center analysis and energy consumption measurements. The adequacy between title and content is excellent.

154 words

Title / Content Match

The title accurately reflects the content: a seminar lecture focused on memory-centric computing.

Quality & Reliability

9/10

Lecture by a leading expert in computer architecture, with references to peer-reviewed papers and technical reports. The content is well-structured and based on extensive research, though it is a lecture rather than a peer-reviewed publication.

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 the lecture, providing an overview of processing-in-memory.
Memory-Centric Computing: Solving Computing's Memory Problem — Recommended reading, discussing the memory problem and solutions.
Memory-Centric Computing: Recent Advances in Processing-in-DRAM — Recommended reading, covering recent advances in processing-in-DRAM.
Intelligent Architectures for Intelligent Computing Systems — Recommended reading, discussing intelligent architectures.
RowHammer: A Retrospective — Recommended reading, providing a retrospective on RowHammer.
Fundamentally Understanding and Solving RowHammer — Recommended reading, discussing RowHammer solutions.
Accelerating Genome Analysis via Algorithm-Architecture Co-Design — Recommended reading, discussing genome analysis acceleration.
From Molecules to Genomic Variations: Accelerating Genome Analysis via Intelligent Algorithms and Architectures — Recommended reading, discussing intelligent genome analysis.

Concurring Sources

A Modern Primer on Processing in Memory — Provides a comprehensive overview of PIM, aligning with the lecture's themes.
Memory-Centric Computing: Solving Computing's Memory Problem — Directly addresses the memory problem and solutions, supporting the lecture's arguments.

External References

Contribution & Novelties

The lecture provides a comprehensive overview of memory-centric computing, synthesizing recent research and highlighting the fundamental inefficiencies of current architectures. It offers a clear framework for understanding the memory bottleneck and proposes a paradigm shift towards computation near data. The talk is valuable for researchers and students in computer architecture, providing both motivation and direction for future work.

Pour aller plus loin :

Processing-in-Memory — Overview of the concept and its history.
RowHammer — Explanation of the DRAM disturbance issue.
Amdahl’s Law — Relevant to the discussion of bottlenecks and acceleration.

91 words

Radar Profile

The radar profile shows high scores across all dimensions, indicating a well-rounded and authoritative lecture. The high scores in quantity and quality of information reflect the depth and rigor of the content, while the technical level is appropriate for an advanced audience. The overall reliability is high, consistent with the lecturer's expertise and the cited sources.

Reliability 9/10

💬 No comments were provided for analysis.