Neuromorphic computers master physics simulations

Overview

For decades, simulating the physics of airplane wings, nuclear weapons, or weather systems required warehouse-sized supercomputers consuming megawatts of power. Researchers at Sandia National Laboratories have now demonstrated that brain-inspired neuromorphic chips can solve these same equations—the partial differential equations underlying nearly all physics simulations—with a fraction of the energy.

The breakthrough, published in Nature Machine Intelligence, arrives as data centers' electricity consumption threatens to double by 2026, driven largely by artificial intelligence. Neuromorphic computing, which mimics how biological neurons communicate through discrete electrical spikes rather than continuous signals, could enable complex scientific simulations on devices small enough to fit inside a drone or satellite—or scale up to replace the most energy-hungry machines in the nuclear weapons complex.

Play on this story Voices Debate Predict

Key Indicators

1.15B

Artificial neurons

Number of neurons in Intel's Hala Point system, now operational at Sandia—roughly equivalent to an owl's brain.

99%

Parallelizable

The NeuroFEM algorithm's efficiency in distributing computations across neuromorphic cores.

12 years

Hidden connection

How long the link between cortical network models and partial differential equations went unrecognized.

2,600W

Hala Point power

Maximum power consumption for Sandia's neuromorphic system—compared to 13,000,000W for a typical supercomputer.

Interactive

Exploring all sides of a story is often best achieved with Play.

Ever wondered what historical figures would say about today's headlines?

People Involved

Brad Aimone

Leading neuromorphic computing research for national security applications

Brad Theilman

Developed NeuroFEM algorithm for physics simulations

Organizations Involved

Sandia National Laboratories

Federal Research Laboratory

Operating world's largest neuromorphic systems for national security research

Department of Energy laboratory responsible for nuclear weapons engineering and national security science.

Intel Corporation

Incumbent x86 chipmaker

Building world's largest neuromorphic hardware platforms

Semiconductor manufacturer developing neuromorphic Loihi processor family since 2017.

International Business Machines Corporation (IBM)

Technology and Consulting Corporation

Pioneered neuromorphic computing with TrueNorth and NorthPole chips

Technology company that developed TrueNorth neuromorphic chip under DARPA funding.

Timeline

Breakthrough Gains Wider Recognition
Publication

Science news outlets report on Sandia's demonstration that neuromorphic computers can perform physics simulations previously requiring energy-intensive supercomputers.
February 14th, 2026
NeuroFEM Paper Published
Research

Theilman and Aimone publish "Solving sparse finite element problems on neuromorphic hardware" in Nature Machine Intelligence, demonstrating that spiking neural networks can solve partial differential equations with 99% parallelizability.
January 7th, 2025
Sandia Partners with SpiNNcloud
Partnership

Sandia announces collaboration with German startup SpiNNcloud to explore neuromorphic computing for nuclear deterrence applications.
May 8th, 2024
Hala Point Arrives at Sandia
Infrastructure

Intel delivers Hala Point neuromorphic system to Sandia—1,152 Loihi 2 chips with 1.15 billion neurons in a chassis the size of a microwave, supporting 20 petaops at 2,600 watts maximum.
April 17th, 2024
Sandia Team Wins International Neuromorphic Prize
Recognition

Brad Aimone leads Sandia team to international prize for demonstrating neuromorphic solutions across heat transfer, medical imaging, and financial problems.
December 1st, 2023
IBM Unveils NorthPole Chip
Hardware

IBM releases NorthPole neuromorphic processor optimized for 2-, 4-, and 8-bit precision, outperforming conventional architectures on image recognition at reduced energy cost.
October 1st, 2023
Intel Launches Loihi 2
Hardware

Intel releases second-generation Loihi chip with improved performance and efficiency, enabling larger-scale neuromorphic systems.
September 1st, 2021
Intel Releases First Loihi Chip
Hardware

Intel unveils Loihi neuromorphic processor with 131,072 artificial neurons and 130 million synapses, built on 14-nanometer technology.
September 1st, 2017
IBM Unveils TrueNorth Chip
Hardware

IBM releases TrueNorth neuromorphic processor with 1 million neurons, 256 million synapses, and 5.4 billion transistors—running on just 70 milliwatts of power.
August 1st, 2014
Balanced Excitation-Inhibition Model Published
Research

Neuroscientists publish cortical network model balancing excitatory and inhibitory signals—the algorithm Sandia would later link to physics equations 12 years later.
January 1st, 2013
DARPA Launches SyNAPSE Program
Funding

The Defense Advanced Research Projects Agency (DARPA) awards $10.8 million to IBM and HRL Laboratories to develop brain-inspired computing hardware, ultimately investing $52 million through 2014.
November 1st, 2008
Neuromorphic Computing Concept Emerges
Research

Carver Mead and Misha Mahowald at Caltech develop first silicon retina and artificial neurons, establishing neuromorphic engineering as a field.
January 1st, 1980

Scenarios

Predict which scenario wins. Contrarian picks score more — points lock in when the scenario resolves.

Predictions leaderboard

Neuromorphic Systems Enter Production for Scientific Computing

Within 3-5 years, neuromorphic systems become standard tools for specific physics simulation tasks at national laboratories. Early adoption focuses on problems well-suited to the architecture—fluid dynamics, structural mechanics, electromagnetic fields—where energy efficiency matters more than raw precision. Supercomputers remain essential for the highest-fidelity simulations, but neuromorphic chips handle exploratory runs and edge deployments.

Discussed by: Intel research division, Department of Energy Advanced Simulation and Computing program, IEEE Spectrum

Consensus —

Neuromorphic Supercomputer Achieves Parity with Traditional HPC

A decade-scale effort produces the first neuromorphic system capable of full-scale nuclear stockpile simulations, reducing the energy footprint of the weapons complex's computing infrastructure by an order of magnitude. This requires solving harder PDEs (nonlinear, coupled systems) and achieving numerical precision comparable to current supercomputers—both active research challenges.

Discussed by: Sandia National Laboratories, The Register, Interesting Engineering

Consensus —

Edge Deployment Transforms Autonomous Systems

Neuromorphic physics simulation enables autonomous drones, satellites, and vehicles to run real-time environmental models onboard rather than relying on pre-computed data or cloud connections. A drone could model airflow around obstacles; a satellite could simulate atmospheric conditions. The low power consumption makes this feasible for battery-powered systems.

Discussed by: IEEE Spectrum, Prophesee, autonomous vehicle researchers

Consensus —

Technical Barriers Limit Scope to Niche Applications

Neuromorphic systems prove excellent at specific problem classes but cannot generalize to the full range of PDEs required for comprehensive scientific simulation. Precision limitations, difficulty with nonlinear equations, and the need for specialized algorithm development for each problem type confine the technology to narrow applications while traditional computing continues to dominate.

Discussed by: HPCwire, academic critics of neuromorphic computing

Consensus —

Historical Context

DARPA SyNAPSE Program (2008-2014)

November 2008 - August 2014

What Happened

DARPA invested $52 million to develop brain-inspired computing hardware, partnering with IBM, HRL Laboratories, and universities. IBM's team built successive prototypes, culminating in the TrueNorth chip with 1 million neurons on a single die consuming just 70 milliwatts—about one ten-thousandth the power density of conventional processors.

Outcome

Short Term

TrueNorth demonstrated that neuromorphic hardware could achieve radically better energy efficiency for specific tasks like image recognition.

Long Term

The program established neuromorphic computing as a serious research field, leading Intel, Samsung, and others to launch competing efforts.

Why It's Relevant Today

Sandia's NeuroFEM breakthrough builds directly on hardware that descended from SyNAPSE—Intel's Loihi architecture emerged partly in response to TrueNorth's success.

ENIAC and the Birth of Scientific Computing (1945)

February 1946

What Happened

The Electronic Numerical Integrator and Computer (ENIAC) began operation at the University of Pennsylvania, consuming 150 kilowatts to perform calculations for hydrogen bomb design. It replaced months of human computation with hours of machine work, but required its own electrical substation.

Outcome

Short Term

ENIAC proved electronic computers could solve physics problems far faster than any alternative, launching the era of computational science.

Long Term

Scientific computing scaled up over 80 years to exascale supercomputers—but energy consumption scaled with it, creating today's data center power crisis.

Why It's Relevant Today

Neuromorphic computing represents a potential architectural break from the von Neumann paradigm that has dominated since ENIAC. If successful, it would be the first fundamental shift in how we compute physics since 1945.

GPU Computing Revolution (2006-2012)

2006-2012

What Happened

Researchers discovered that graphics processing units, designed for video games, could accelerate scientific simulations and machine learning by 10-100x compared to conventional processors. NVIDIA released CUDA in 2006; by 2012, GPUs dominated high-performance computing for certain workloads.

Outcome

Short Term

GPUs enabled the deep learning revolution by making neural network training practical at scale.

Long Term

GPU computing became the default architecture for AI, but inherited the energy efficiency limitations of conventional silicon—leading to today's power consumption concerns.

Why It's Relevant Today

Neuromorphic computing could be the next architectural wave after GPUs. Just as GPUs found unexpected applications beyond graphics, neuromorphic systems designed for AI inference may transform scientific computing.