Review of Applications in Hardware-Aware AI Research for High Energy Physics
The DOE Office of Science program in High Energy Physics intends to hold a review of applications through the FY 2024 Continuation of Solicitation for the Office of Science Financial Assistance Program from eligible applicants, such as Universities, in Hardware-Aware AI Research for High Energy Physics. Applications are sought in the technical areas described in the Hardware-Aware Artificial Intelligence for High Energy Physics Lab 24-3305. Eligibility and application requirements can be found in DE-FOA-0003177. Answers to Frequently Asked Questions can be found here.
For full consideration applications should be received by July 24th, 2024. Investigators planning to write applications in this subject matter are requested to submit pre-applications in the Portfolio Analysis and Management System (PAMS) at https://pamspublic.science.energy.gov by June 26th, 2024.
HEP plans to commit approximately $15 million in current and future fiscal year funds to support awards in hardware-aware AI. Awards are expected to range from $100,000 per year to $350,000 per year, with median award size of $250,000 per year, with an expected duration of three years. Any grants are expected to be made after October 1, 2024.
An informational webinar was held May 29th at 1pm ET, to provide additional information about the review and answer questions from the community. (Slides | Video)
Program Summary
DOE Office of Science program in High Energy Physics invites proposals for AI research that requires and relies on HEP hardware systems primarily in two topic areas: Smart Detectors, defined as intelligence on detector in measurement and readout and control electronics; and AI for Operations, for real-time facility, experiment and observatory operations and control. The HEP program explicitly encourages proposals that include partnerships with non-traditional HEP institutions and researchers that may broaden the participation in HEP research and the AI sub-field.
Program Objective
Resulting awards will invest in AI research applied to HEP hardware systems. Hardware systems for this Announcement refers to HEP specific detector and sensor technologies deployed in HEP experiments and facilities or under development for future HEP applications including Application Specific Integrated Circuits (ASICs) and readout electronics that provide real-time operation of facilities, experiments, and observatories. Widely used computational hardware such as CPUs, GPUs, FPGAs, and Quantum Processors as well as emulations of those systems are not included in the definition of Hardware systems for this Announcement. This opportunity follows previously identified national research priorities and Basic Research Needs workshops conducted by SC and HEP [1, 2, 3, 4] and follows previous support through the Data, Artificial Intelligence, and Machine Learning at DOE Scientific User Facilities (LAB 20-2261) and Artificial Intelligence Research for High Energy Physics (DE-FOA-0002705). The objectives are to support AI research that extends the scientific potential of existing and future HEP hardware systems well beyond what is currently achievable both in measurement and readout capability and in operations and control. Applications are sought that make use of and develop cutting edge AI technologies and techniques for use in HEP and that develop tools and datasets to enable broader participation in the HEP AI Ecosystem.
Smart Detectors: HEP detectors rely on sensors and ASICs and custom electronics to configure and control the sensing elements as well as to read out and process the signals produced. These systems often requiring analog and digital circuits based on complex functional requirements (such as power regulation, high speed communications, data buffering, clock multiplication, radiation hardness, etc.). They make use of pre-designed intellectual property (IP) blocks in the target technology and often must operate in extreme environmental conditions, potentially with limited access for maintenance. AI research into algorithms or IP development that improves the ability of HEP specific ASICs to meet functional requirements and improve operational robustness is sought. This may include AI applications that do not use traditional ML such as those relying on detector sense element response, analog circuits, or distributed and emergent intelligence.
Proposals are sought to develop novel triggering and readout schemes beyond traditional FPGA/GPU/CPU systems with centralized decision making. Proposals for ML-based triggering algorithms in traditional hardware architectures are not encouraged.
AI for Operations: AI approaches can have a significant impact on increasing the productivity of complex scientific facilities and HEP experiments. They can improve: (i) the efficiency by minimizing down time, automating control of complex systems, and reducing time to optimal conditions; and (ii) the quality of the data recorded through enhanced monitoring, lowering the level of expertise needed to understand and perform complex actions, and reduce the level of human intervention necessary.
AI techniques that are robust against unpredictable or unanticipated and changing conditions, that may occur over varying timescales, while still achieving the above objectives are especially encouraged. Systems that make the process of data curation and collection uptime more robust against “on shift” human operator error will be considered. ML techniques, such as digital twin or surrogate models, may be applied to real-time systems where traditional simulation is not currently possible. Proposals targeting real-time systems that may be used to monitor and predict beam delivery, or experiment or observatory conditions to potentially intervene and change data-taking or operating modalities are encouraged.
Topics related to quantum computing and quantum machine learning are not responsive, whereas AI research that targets quantum experiments and control systems may be. ML-based simulation of HEP hardware and facilities for offline systems are not responsive.
References
