U.S. flag

An official website of the United States government

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.

Information Technology /Artificial intelligence

Hardware for AI

Artificial intelligence (AI) applications are becoming more and more prevalent in our everyday lives. Most contemporary implementations of AI use digital logic and the conventional CMOS hardware that has enabled the information revolution. A team of NIST researchers seeks to enable future generations of AI by focusing on fabricating and measuring new brain-inspired circuits and architectures based on novel devices to deliver computing methods, speeds, and energies better than those achievable using the current computing paradigm. Conventional computing represents information with binary encoding – ones and zeros times different powers of two. The variety of approaches studied at NIST are founded on the concept that computing can be more efficient when information is directly represented by the physical properties of devices themselves. The devices then perform the computation directly, as opposed to the conventional approach of manipulating the binary representation.

The need for these novel approaches is driven by several realities of modern computing. We are pushing computers to take on tasks that humans are much better at than traditional computers. The demands for this type of computing is growing much faster than the capabilities of traditional computers. Perhaps most of all, the energy required to deliver the computations is the most rapidly increasing sector of energy consumption in the world, and it must be reduced to limit the impact on the climate. It is also essential to make computing more efficient in “edge” applications in which computers are embedded in devices that have very restricted energy supplies. The efficiency of the brain drives research that identifies devices acting like the neurons and synapses of the brain and uses them to enable algorithms that compute like the brain. NIST’s AI Hardware team’s research aims to develop the necessary device-level and circuit-level measurements and theory to support the evolution of this technology from laboratory research to commercial application.

The Research

Projects & Programs

circuit

Emerging Hardware for Artificial Intelligence

Ongoing
Here is a brief description of our work with links to recent papers from our investigations, broadly classified as experimental and modeling. A brief overview of Josephson junction-based bio-inspired computing can be found in our review article. Experimental We have facilities to develop our devices
He4 cryostat

Physics and Hardware for Intelligence

Ongoing
Our work in this area can be separated into two categories: conceptual and experimental. Please read our publications linked below for more information. Experimental: Our latest generation of synaptic circuits are described in a 2024 paper published in APL Machine Learning. These circuits are our
Fig. 1. Die containing 285 superparamagnetic tunnel junctions.

Spintronics for Neuromorphic Computing

Ongoing
Magnetic tunnel junctions (see Fig. 1) consist of two thin films of ferromagnetic material separated by a few atomic layers of an insulating material. The insulator is so thin that electrons can tunnel quantum mechanically through it. The rate at which the electrons tunnel is affected by the
Small array of transistors designed for single device integration.

Integrated CMOS Testbeds for Nanoelectronics and Machine Learning

Ongoing
The increasingly complex device requirements for next-generation computing architectures such as neuromorphic computing or nanoelectronic machine learning accelerators present challenges for researchers across the spectrum of institutions, from small businesses and universities to government
Fig. 1. Time re-Envisioned: Not a biproduct of computation but the domain in which information is encoded. Leads to different mathematical functions becoming energy efficient primitives.

Temporal Computing

Ongoing
In standard integrated circuits, information that is coded as ones and zeros is implemented by voltages on wires being high or low. The circuits consume energy during transitions between these voltages. Binary numbers have a voltage per bit so there are a lot of transitions each time a number
EBIC Current

Neuromorphic Device Measurements

Ongoing
One type of device that is emerging as an attractive artificial synapse is the resistive switch, or memristor. These devices, which usually consist of a thin layer of oxide between two electrodes, have conductivity that depends on their history of applied voltage, and thus have highly nonlinear
View All Projects/Programs
Illustration with colored cylindrical flasks on the left, arrays of bar magnets in the center, and bottles of wine on the right.

Nanomagnets Can Choose a Wine, and Could Slake AI’s Thirst for Energy

By analyzing the different characteristics of wines, such as acidity, fruitiness and bitterness (represented as colored flasks on the left), a novel AI system (center) successfully determined which type of wine it was (right). The AI system is based on magnetic devices known as "magnetic tunnel junctions," and was designed and built by researchers at NIST, the University of Maryland and Western Digital.

Read More