The spotlight on large models falls heavily on algorithms and products, but the physical reality underpinning them is rarely discussed: behind every conversation lies a data center humming with GPUs, every watt of power costs real money, and the demand for running AI on phones, cars, and watches is turning “how to make neural networks smaller, faster, and more energy-efficient” into the industry’s critical bottleneck. NVIDIA’s market cap mythos is merely the most conspicuous footnote to this proposition—beneath it lies an entire field of AI hardware and systems research, solving the question of “where computing power comes from and where energy goes” day in and day out for decades.
Professor Yiran Chen is one of the representative figures among Chinese scholars in this field. If algorithm researchers are teaching AI to think, his school of researchers is building a more efficient “brain” for AI.
Who Is Yiran Chen?
Yiran Chen is a Professor in the Department of Electrical and Computer Engineering at Duke University, an IEEE Fellow and ACM Fellow. He has long served as the Director of Duke’s Center for Computational Evolutionary Intelligence (CEI) and led large-scale NSF-funded research projects on edge computing. He earned his undergraduate degree from Tsinghua University and his Ph.D. from Purdue University. His academic career spans both industry (where he conducted storage research at Seagate and HP Labs) and academia—an industrial experience in the storage sector that profoundly shaped his later research trajectory of “rethinking computation from a storage perspective.”
He also holds significant public influence within Chinese tech communities: active on platforms such as Zhihu, he frequently offers accessible insights on AI academia, the chip industry, and studying abroad for research. He is one of the few top-tier Chinese scholars willing to consistently produce serious popular science content for a Chinese-speaking audience—giving Chinese readers another entry point to “follow Yiran Chen” beyond his academic papers.
Main Research Directions and Contributions
Neural Network Compression and Efficient AI
Deep models often boast billions of parameters, while the computing power and battery life of phones and embedded devices are immutable constraints. Professor Chen’s team work in model compression covers the three core pillars of this field:
Pruning: Removing redundant weights and neurons from the network to slim down the model with minimal loss in accuracy—the team’s achievements in structured pruning have been widely cited.
Quantization: Compressing parameters from 32-bit floating point to integers below 8 bits, halving memory usage and computational load.
Knowledge Distillation: Using a large model as a teacher and a small model as a student to “transfer” capabilities to the lightweight model.
The ability of today’s phones to run voice assistants or cameras to perform real-time AI processing is the result of two decades of accumulation in this entire research direction—it may not be sexy, but it is ubiquitous.
Processing-in-Memory (PIM)
In traditional von Neumann architectures, data shuttles back and forth between memory and processors; the energy cost of this movement far exceeds that of computation itself—this is the famous “memory wall,” which hits data-intensive workloads like AI particularly hard. The logic behind processing-in-memory is to address the root cause: allowing computation to occur directly within the memory.
Professor Chen is an international pioneer in using novel non-volatile memories (such as RRAM memristors) for neural network acceleration; his team’s early work in this direction is considered foundational literature in the field. This route has now moved from academic concepts to industrial trials and is viewed as a key candidate answer for AI computing power in the post-Moore’s Law era.
Robustness and Security of AI Systems
Adversarial sample attacks, model reliability under noise and faults—once AI enters life-critical scenarios such as autonomous driving and healthcare, these issues shift from academic curiosities to engineering red lines. This is also a key direction for his team.
Academic Impact
With hundreds of papers in top conferences and journals spanning circuits (ISSCC), EDA (DAC/ICCAD/DATE), and machine learning (NeurIPS/ICML)—this cross-community publication record itself illustrates the interdisciplinary nature of “hardware and AI.” He has received multiple best paper awards, boasts tens of thousands of citations, and has mentored a large cohort of students now active in both academia and the chip industry, with many holding faculty positions at top universities. Another measure of academic impact is “academic descendants,” an area where he excels.
Learning Resources
Papers: Search “Yiran Chen Duke” on Google Scholar; review articles on model compression and processing-in-memory serve as excellent introductory maps.
Chinese Content: Search for his account on Zhihu. His answers and articles on AI chips, research paths, and industry assessments provide a far more accessible Chinese entry point than papers—particularly valuable for student readers regarding “how to do research” and “how to choose a direction.”
Lectures: Keynotes from major conferences and university lectures are available on YouTube and Bilibili, primarily focusing on edge AI and efficient computing.
Who Should Follow His Work
Researchers and Students in AI Chips/Architecture: An indispensable name in this field; his papers and reviews are standard references.
Engineers Working on Model Deployment Optimization: Pruning, quantization, and distillation are used daily; understanding the technical lineage strengthens engineering intuition.
Professionals Focused on Edge AI and On-Device Large Models: Running large models on phones and cars is one of the hottest current industry directions, and his research sits at the theoretical forefront of this track.
Students Planning to Apply for Overseas AI Hardware Programs: Duke’s CEI is a top-tier team in this direction, and Professor Chen’s Chinese popular science content offers a rare window into the US research ecosystem.
Note
To be direct: this is an academic profile, not a registrable product. If you are looking for “usable AI tools,” this page will not help.
Its value lies in another dimension: when everyone is talking about models and applications, understanding the “physical layer of AI”—computing power, energy consumption, chips—is a required course for grasping the depth of this industry. Professor Chen’s research career happens to be one of the best living textbooks for this course: from storage to computation, from the lab to industry, the other half of the AI revolution is written in hardware.