Berkeley Lab has been employed artificial intelligence and machine learning in its research for decades. Following is a sampling of patented, patent-pending, and copyrighted technologies that have leveraged AI/machine learning:
GRID AND ENERGY MANAGEMENT
Smart Control of Distributed Energy Resources
This Modelica model library was developed to facilitate the simulation and optimization of distributed energy resources such as photovoltaics, battery storage, smart inverters, electric vehicles, and electric power systems.
Developers: Christoph Gehbauer, J. Müller, T. Swenson
Contact: Michael Wetter, MWetter@lbl.gov
Distribution Substation Planning Toolkit (dsp-toolkit) 2025-051
The Distribution Substation Planning Toolkit (DSP Toolkit) is a software suite designed to streamline the planning and optimization of distribution substations. This toolkit offers a comprehensive set of tools and APIs for data curation, short-term electric load forecasting, and weather-sensitive load adjustment, making it an resource for utility companies, engineers, and researchers.
Developers: Miguel Heleno, Tianzhen Hong, Luis Rodriguez Garcia, Wanni Zhang, Kaiyu Sun, Han Li
Contact: Tianzhen Hong, thong@lbl.gov
EnergyPlus Model Context Protocol Server (EnergyPlus-MCP) 2025-203
EnergyPlus-MCP is the first open-source Model Context Protocol server specifically designed for EnergyPlus building energy simulation. This software enables AI assistants and other applications to interact programmatically with EnergyPlus through a standardized, secure interface, eliminating traditional technical barriers in building energy modeling.
Developers: Tianzhen Hong, Yujie Xu, Han Li
Contact: Tianzhen Hong, thong@lbl.gov
COMPUTATION
FPGA-based In-Situ Machine Learning for Real-time Quantum State Discrimination 2024-145
This technology enables real-time quantum state discrimination on field programmable gate array (FPGA) hardware by integrating a multi-layer neural network onto an RFSoC platform, eliminating the latency and errors associated with transferring quantum data to host computers for processing, and achieving precise and rapid computation.
Inventors: Yilun Xu, Neel Vora, Gang Huang, Phuc Nguyen
Contact: ipo@lbl.gov
Gradient-Descent Training for Thermodynamic Computers 2025-199
This technology enables ultra–low-energy machine-learning computation by introducing a gradient-descent training method for thermodynamic computers that harness thermal noise to perform neural-network-like operations efficiently out of equilibrium, achieving high accuracy while dramatically reducing energy consumption compared to conventional digital computing.
Inventor: Stephen Whitelam
Contact: Lucian Sweitzer, lsweitzer@lbl.gov
Energy-Efficient Superconducting Tsetlin Machines for Efficient Deep Neural Networks/Machine Learning Applications 2024-078
Scientists at Berkeley Lab have developed a novel implementation of the Tsetlin machine using superconducting rapid single-flux quantum (RSFQ) technology. This innovation combines the interpretability and efficiency of Tsetlin machines with the ultra-low power consumption and high processing speeds of superconducting circuits, potentially revolutionizing energy-efficient, high-speed computing for machine learning applications.
Inventors: Dilip Vasudevan, Ran Cheng, Christoph Kirst
Contact: ipo@lbl.gov
INDUSTRIAL TECHS (e.g., LASER, MACHINE VISION, OPTICAL SORTING)
Spectral kernel machines with electrically tunable photodetectors 2026-037
This technology enables real-time, ultra-low-power spectral intelligence by performing analog machine-learning inference directly during light detection with a tunable photodetector, eliminating data bottlenecks and enabling fast, high-accuracy material and object identification for sensing, metrology, and remote-platform applications.
Inventors: Ali Javey, Dehui Zhang, Aydogan Ozcan, Yuhang Li
Contact: Lucian Sweitzer, lsweitzer@lbl.gov
Machine-learning Based Stabilization Controller for Complex or Unstable Environments 2021-133
This technology provides a fast, adaptive machine-learning controller that stabilizes complex multi-input/multi-output systems in drifting or unstable environments by continuously learning differential system behavior without requiring a full physical model or system-wide retraining, enabling real-time control for applications such as laser stabilization, robotics, and industrial automation.
Inventors: Dan Wang, Qiang Du, Russell Wilcox, Tong Zhou, Christos Bakalis, Derun Li
Contact: Lucian Sweitzer, lsweitzer@lbl.gov
Advanced Pulse-to-Pulse Stabilization of Laser via Machine Learning 2025-036
A machine learning approach to correcting laser pointing errors in real time, serving as a robust implementation of predictive control in high-power, low-repetition rate lasers for pulse-to-pulse stabilization. Benefits include unprecedented stabilization accuracy, precise pilot beam, a robust AI engine, and a full integrated system.
Developers: Qiang Du, Dan Wang, Alessio Amodio, Curtis Berger, Anthony Gonsalves, Hai-En Tsai, Samuel Barber, Jeroen van Tilborg, Russell Wilcox, Alex Picksley, Zak Eisentraut, Neel Rajeshbhai Vora, Mahek Logantha, Qing Ji
Contact: Lucian Sweitzer, lsweitzer@lbl.gov
AutoEMX for Identifying Material Phases in Inorganic Powder Samples
AutoEMX is a Python-based software platform that fully automates scanning electron microscopy (SEM)-energy-dispersive X-ray spectroscopy (EDS) measurements and applies machine learning to accurately identify and quantify individual material phases in inorganic powder samples. Unlike conventional SEM-EDS tools, it delivers unprecedented accuracy on closely intermixed phases, enabling fully automated compositional analysis.
Developers/Investigators: Andrea Giunto, Gerd Ceder
Contact: Lucian Sweitzer, lsweitzer@lbl.gov
RADIATION DETECTION
SPECTRE-ML: Spectral Peak Enhancement via Machine Learning 2023-100
SPECTRE-ML is a machine-learning software platform that optimizes clustering of segmented radiation detectors to significantly improve spectroscopic performance by learning detector-specific effects such as dead layers and gain shifts. Designed for highly segmented CZT detectors in IAEA safeguards, it delivers more accurate nuclear material verification and can be readily extended to other next-generation detector systems.
Developers: Jayson Vavrek, Gabriel Aversano, Micah Folsom, Daniel Hellfeld, Hannah Parrilla
Contact: Lucian Sweitzer, lsweitzer@lbl.gov
BIOSYNTHESIS TOOLS AND THERAPEUTIC TECHNOLOGIES
BilboMD Software for Modeling 3D Biological Molecules in Solution 2025-173
BilboMD is a web-based software pipeline that combines X-ray scattering data with macromolecular structure prediction and molecular dynamics simulations to accurately model the three-dimensional structures of biological molecules in solution, as they would exist inside of the organism or body.
Developers: Scott Classen, Michal Hammel
Contact: Jeremy Greeter, jgreeter@lbl.gov
DelPlasmid: Identification of Plasmids with Deep Learning and Machine Learning 2019-037
DelPlasmid helps identify plasmids in assembled bacterial genomes. It is Long Short-Term Memory (LSTM)-based on a deep learning model that takes as input a combination of assembled sequences and extracted features to identify bacterial plasmids. This model was trained on high-quality plasmid sequences from the ACLAME database and the NCBI Refseq.microbial dataset. The tool achieved an AUC-ROC of 91% on a 5-fold cross-validation.
Developers: William Andreopoulos, Jan Balewski
Contact: Jeremy Greeter, jgreeter@lbl.gov
Kinetic Learning, Kinetic Deep Learning Software for Accelerating Strain Engineering 2018-038, 2025-084
This technology uses machine learning models to simulate virtual strains of organisms and identify potential biological modifications. Kinetic Learning software (2018-038) enables rapid, virtual simulation of engineered biological strains by using kinetic machine-learning models to predict how genetic or pathway modifications will affect organism behavior, reducing costly trial-and-error experimentation in synthetic biology. Kinetic Deep Learning software (2025-084) uses deep learning to accurately predict time-series metabolite concentrations directly from protein-level data, replacing traditional kinetic equations with data-driven pathway modeling to reliably design and optimize bioengineered production strains across diverse hosts and products.
Developers: Hector Garcia Martin, Patrick Kinnunen, Tijana Radivojevic, Jose Manuel Marti, Zachary Costello
Contact: Robin Johnston, rjohnston@lbl.gov
AI-Assisted Prediction of Phage-Host Absorption Factors 2025-150
This technology uses a machine-learning platform trained on genome-wide functional screens and phage genome sequences. The technology rapidly and accurately predicts bacteriophage host receptors, enabling fast, large-scale identification of phage–bacteria interactions to accelerate phage therapy, receptor discovery, and synthetic biology applications.
Inventors: Lucas Moriniere, Avery Noonan, Adam Arkin, Vivek Mutalik
Contact: Jeremy Greeter, jgreeter@lbl.gov
AI-Powered Cellular Morphometric Biomarkers for Classification of Prostate Cancer for Neoadjuvant Androgen Deprivation Therapy (NADT) 2025-013
This technology uses an AI-based analysis of prostate cancer biopsy samples to identify cellular morphometric biomarkers that predict patient response to therapy, enabling personalized treatment selection that improves outcomes, reduces unnecessary therapy, and identifies alternative drug sensitivities.
Inventors: Hang Chang, April Mao, Jian-Hua Mao
Contact: Jeremy Greeter, jgreeter@lbl.gov
ART: A Machine Learning Automated Recommendation Tool for Guiding Synthetic Biology (2020-011)
The patent-pending Automated Recommendation Tool (ART) uses probabilistic modeling techniques to guide metabolic engineering systematically without requiring a full mechanistic understanding of the biological system. Using sampling-based optimization, ART provides a set of recommended strains to be built in the next engineering cycle, alongside probabilistic predictions of their production levels. Demonstrated to have high predictive accuracy. Using ART improved tryptophan titer and productivity by up to 74 and 43%, respectively, compared to the best designs used for algorithm training.
Developers: Hector Garcia Martin, Zachary Costello, Tijana Radivojevic
Contact: Robin Johnston, rjohnston@lbl.gov
