BEGIN:VCALENDAR VERSION:2.0 PRODID:Linklings LLC BEGIN:VTIMEZONE TZID:America/Los_Angeles X-LIC-LOCATION:America/Los_Angeles BEGIN:DAYLIGHT TZOFFSETFROM:-0800 TZOFFSETTO:-0700 TZNAME:PDT DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0700 TZOFFSETTO:-0800 TZNAME:PST DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260402T024533Z LOCATION:3003\, Level 3 DTSTART;TZID=America/Los_Angeles:20250622T133000 DTEND;TZID=America/Los_Angeles:20250622T170000 UID:dac_DAC 2025_sess250_WKSHP017@linklings.com SUMMARY:First International Workshop on Synergizing AI and Circuit-System Simulation DESCRIPTION:Zhengqi Gao and Duane Boning (Massachusetts Institute of Techn ology), Zhou Jin (Zhejiang University), Yanqing Zhang and Haoxing Ren (Nvi dia), Yiran Chen (Duke University), and Ron Rohrer (CMU)\n\nHardware tape- outs are prohibitively expensive and time-consuming, making circuit and sy stem (CAS) simulators crucial for verifying designs efficiently and cost-e ffectively prior to fabrication. An extensive array of simulators exists t oday, tailored for various CAS applications, such as Verilog simulators fo r digital integrated circuits (ICs), SPICE-based simulators for analog ICs , Verilog-AMS simulators for mixed-signal systems, and electromagnetic sim ulators for high-frequency circuits and antennas. Despite decades of devel opment and the high degree of maturity achieved by CAS simulators, the rec ent surge of artificial intelligence (AI) is rekindling renewed interest f rom both software and hardware perspectives. On the hardware front, the ex ceptional parallelism capabilities of GPUs can be harnessed to expedite CA S simulations, such as GPU-accelerated SPICE simulations and logic gate si mulation. On the software side, deep learning (DL) algorithms are being se amlessly integrated into CAS simulators serving as surrogate models or pro viding initial guesses, to reduce computational workloads and improve eff iciency. Conversely, the principles of CAS simulation are catalyzing novel AI models. One prominent example is the use of ordinary differential equa tions (ODEs), which have long been a cornerstone of time-domain analog cir cuit simulations in SPICE, with the adjoint method used for gradient compu tations. In the DL community, these techniques have evolved into Neural OD Es, a class of models that parameterize ODE dynamics using neural networks . Neural ODEs have proven especially effective for time-series forecasting and are closely linked to the development of generative diffusion models. Similarly, state-space models (SSMs), once the bedrock of linear time-inv ariant systems, now underpin architectures such as Mamba, designed for eff icient natural language processing. Another notable adaptation of classica l circuit principles in modern AI is Kirchhoff's current law (KCL), which has been leveraged to construct analog neural networks, such as memristor crossbar arrays and KirchhoffNet. Furthermore, Fourier transforms, widely used in frequency-domain CAS simulations for signal processing, have been reimagined as neural operators. This adaptation has led to breakthroughs i n AI-driven scientific applications, such as weather forecasting.\n\nThe s imilarities between CAS simulation and AI are profound, yet no dedicated p latform exists for researchers, engineers, and practitioners to discuss th is interdisciplinary topic. Recognizing this critical need, the First Inte rnational Workshop on Synergizing AI and Circuit-System Simulation aims to bring together experts to explore innovative methodologies that leverage the synergies between these fields. The workshop will provide a platform t o discuss recent advancements and foster interdisciplinary collaboration.\ n\nThe workshop contains 5 talks; each is scheduled to be 45 mins.\n\nTitl e: GPU Accelerated Simulation: From RTL to Gate-Level, From Opportunities to Success\nContributors: Yanqing Zhang, Mark (Haoxing) Ren, Nvidia\nAbstr act: In this talk, we will present a brief history of accelerated simulati on, and motivate why GPUs can be an attractive platform to accelerate this uber-important EDA application. We will go through several important type s of simulation abstraction levels: RTL, gate-level, and re-simulation, as well as the unique challenges each type of simulation faces when attempti ng to accelerate them. Next, we go into detailed discussion on some recent research work that aims to attack these challenges, centered around 3 pro jects GEM (GPU accelerated RTL simulation), GL0AM (GPU accelerated gate-le vel simulation), and GATSPI (GPU accelerated re-simulation). Finally, we p rovide some analysis and insight into where the remaining opportunities fo r improvement and research lie (and why), and which challenges have been s uccessfully conquered.\n\nTitle: AI on functions\nContributors: Kamyar Azi zzadenesheli, Nvidia\nAbstract: Artificial intelligence is rapidly advanci ng, with neural networks powering breakthroughs in computer vision and nat ural language processing. Yet, many scientific and engineering challenges— such as material science, climate modeling, and quantum chemistry—rely on data that are not words or images, but functions. Traditional neural netwo rks are not equipped to handle these functional data.\nTo overcome this li mitation, we introduce neural operators, a new paradigm in AI that general ize neural networks to learn mappings between function spaces. Neural oper ators enable AI to process and reason about functional data directly, open ing new frontiers for scientific discovery and technological innovation ac ross diverse disciplines.\n\nTitle: Machine Learning for EDA, or EDA for M achine Learning?\nContributors: Zheng Zhang, University of California at S anta Barbara\nAbstract: The rapid advancement of machine learning (especia lly deep learning) in the past decade has impacted, both positively and ne gatively, many research fields. Driven by the great success of machine lea rning in image and speech domains, there have been increasing interests in “Machine Learning for EDA”. In the first part of the talk, I will explain the main challenge of data sparsity when applying existing machine learni ng techniques to EDA. Then I will show how some data-efficient scientific machine learning techniques, specifically uncertainty quantification and p hysics-constraint operator learning, can be utilized to build high-fidelit y surrogate models for variability analysis and for 3D-IC thermal analysis , respectively. These techniques can greatly reduce the number of required device/circuit simulation data samples. Another important but highly igno red direction is “EDA for Machine Learning”. The five decades of EDA resea rch has produced a huge body of solid theory and efficient algorithms for analyzing, modeling and optimizing complex electronic systems. Many of th e white-box EDA ideas may be leveraged to solve black-box AI problems. In the second part of the talk, I will show how the self-healing idea and com pact modeling idea from EDA can be utilized to improve the trustworthiness and sustainability of deep learning models (including large-language mode ls).\n\nTitle: Optimization Meets Circuit Simulation\nContributors: Aayush ya Agarwal, Larry Pileggi, Carnegie Mellon University\nAbstract: Optimizat ion is central to the design and analysis of modern engineering systems. B ut as systems scale in complexity, traditional optimization tools, which a re often rooted in purely mathematical representations, can struggle to re liably find feasible solutions. In this talk, we explore a new approach th at bridges mathematical optimization with circuit simulation. This approac h maps optimization problems as analog circuits, where optimization compon ents are modeled as equivalent circuit devices connected through a network . This reframes the development of optimization algorithm as the design an d simulation of circuits, which allows us to leverage principles from line ar networks, nonlinear device physics, and solution techniques in SPICE an d its many derivations. The result is a class of physics-inspired methods tailored to the nonlinearities and structure of each optimization problem that would be far less intuitive without the view through a circuit model lens. We demonstrate the efficacy of the equivalent circuit methods for re al-world applications, including training machine-learning, and optimizing power grids.\n\nTitle: Oscillator Ising Machines: Principles to Working H ardware\nContributors: Jaijeet Roychowdhury, University of California at B erkeley\nAbstract: Modern society has become increasingly reliant on rapid and routine solution of hard discrete optimization problems. Over the pas t decade, fascinating analog hardware approaches have arisen that combine principles of physics and computer science with optical, electronic and qu antum engineering to solve combinatorial optimization problems in new ways ---these have come to be known as Ising machines. Such approaches leverage analog dynamics and physics to find good solutions of discrete optimizati on problems, potentially with advantages over traditional algorithms. Unde rlying these approaches is the Ising model, a simple but powerful graph fo rmulation with deep historical roots in physics. About eight years ago, we discovered that networks of analog electronic oscillators can solve Ising problems “naturally”. This talk will cover the principles and practical d evelopment of these oscillator Ising machines (OIMs). We will touch upon s pecialized EDA tools for oscillator based systems and note the role of nov el nanodevices. Applied to the MU-MIMO detection problem in modern wireles s communications, OIMs yield near-optimal symbol-error rates (SERs), impro ving over the industrial state of the art by 20x for some scenarios.\n\nTo pics: AI, Sunday Program\n\n END:VEVENT END:VCALENDAR