Neuromorphic Computing and Artificial Intelligence Hardware

Neuromorphic Computing and Artificial Intelligence Hardware is sponsored by NRC Research Associateship Programs. Supports postdoctoral research developing neuromorphic computing hardware using emerging devices to improve AI speed and energy efficiency.

1 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 This Funding Opportunity Announcement (FOA) describes the research objectives of the Air Force Research Laboratory’s (AFRL) university Center of Excellence (COE) in Brain-Derived Neuromorphic Computing with Intelligent Materials, which addresses research objectives of the United States Space Force (USSF) and the United States Air Force (USAF).

A university COE is defined as a joint effort among Air Force Office of Scientific Research (AFOSR), Air Force Research Laboratory Technology Directorates (AFRL TDs), and an outstanding university or team of universities to perform high-priority collaborative research.

This center is a joint project between AFOSR, the Information Directorate (AFRL/RI), and the Materials and Manufacturing AFOSR anticipates making at least one grant award of up to $1,000,000 per year per award, for a maximum of five years. The base period is for three years, which the AFOSR intends to incrementally fund, followed by an option to extend an additional two years provided the COE passes a midterm “go/no-go” review at the 2.

5-year point. By agreement, AFOSR and the participants AFRL/RI and AFRL/RX can ramp-up funding in beginning years or ramp-down funding in the final years to a COE. All funding decisions are at the Government’s discretion and are subject to the availability of funds.

Proposers are highly encouraged to confer with the designated AFOSR program officer as soon as possible. AFOSR will evaluate proposals using a peer review panel and the criteria specified in section E: “Application Review Information. ” While AFOSR reserves the right to select and fund all, some, or none of the proposals, AFOSR anticipates making one grant award under this announcement.

AFOSR will not provide funding for reimbursement of proposal or application costs associated with responding to this FOA. White Papers briefly summarizing the proposing institution’s ideas are highly encouraged but not required.

The AFOSR program officer will coordinate with the sponsoring Information Directorate (AFRL/RI) and Materials and Manufacturing Directorate (AFRL/RX) leads to provide feedback to white papers and will share responsibility for ensuring the success of a COE.

2 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 Proposals may choose to include a data management plan that outlines how samples and data collected in the program will be stored and managed. This includes, but is not restricted to, issues such as: standards for data and metadata collection, content and format, data archiving, database management, and data sharing within and outside the COE.

This is modeled on the National Science Foundation Data Management Plan 1. This COE will focus on the interactions between several technical elements that form the foundations of brain-based neuromorphic models and architectures. > Hyperlinks have been embedded within this document and appear as underlined, and or blue- > colored words in the midst of paragraphs.

The reader may “jump” to the linked section within > this document by “clicking” (CTRL + CLICK, or CLICK). 1. FEDERAL AWARDING AGENCY NAME Air Force Office of Scientific Research 875 North Randolph Street, STE 325, Room 3112 2.

FUNDING OPPORTUNITY TITLE CENTER OF EXCELLENCE (COE): Brain-Derived Neuromorphic Computing with Intelligent 5. CATALOG OF FEDERAL DOMESTIC ASSISTANCE (CFDA) NUMBER 12. 800 Air Force Defense Research Sciences Program A Proposer’s Day will be held virtually on 03 May 2021 for the purpose of facilitating teaming among prospective proposers .

The Government is not responsible for and will not assist with team creation. Advance registration is required at https://community. apan.

org/wg/afosr/w/researchareas/29659/2021-afrl-center-of-excellence-in- brain-derived-neuromorphic-computing-with-intelligent-materials-proposer-s-day/. If requesting a 5-minute slot for an “elevator pitch” presentation to all attendees, you must register by 28 April 2021 . Slots will be reserved on a first-come, first-served basis.

Proposer’s Day presentations are intended for soliciting teaming relationships, are not a prerequisite for responding to this FOA, and will not influence white paper or proposal evaluations. Presenters’ slides will be made publicly accessible at the registration website following the event. > 1https://nsf.

gov/eng/general/ENG_DMP_Policy. pdf 3 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 Pre-proposal inquiries and questions must be received in writing by electronic mail not later than 07 May 2021 at 11:59 PM Eastern Daylight Time (EDT) to be considered. White papers must be submitted electronically at https://community.

apan. org/wg/afosr/p/submitawhitepaper by 01 June 2021 at 11:59 PM Eastern Daylight Time to be considered. White paper evaluation is meant to initially assess the capability of a proposed effort and is NOT a selection process.

White papers can be up to 8 pages in length plus references. White papers should minimally articulate: 1. An initial list of members of the proposed team.

2. The main technical components of the proposed research and how it aligns with the goals of 3. The specific proposed activities to establish a relationship with AFRL/RI, RX, and 4.

The specific plans to educate students/postdoctoral fellows, expose them to research opportunities with AFRL’s Technical Directorates, and ultimately further the interests of The Government will respond to white papers before COB on 01 July 2021 . Proposals must be received electronically through Grants. gov by 16 August 2021 at 11:59 PM Eastern Daylight Time to be considered.

This Center of Excellence is anticipated to extend the research interests of AFRL in the topical area of neuromorphic computing and provide opportunities for a new generation of US scientists and engineers to address United States Space Force (USSF) and United States Air Force (USAF) research needs.

This is a special FOA because it explicitly calls for (a) research in the high- priority Air Force interest areas of neuroscience, neuromorphic computing, and nanomaterials; and (b) education of students within the US in vital technology areas with opportunities for potential recruitment of US nationals for employment at AFRL. In conjunction with AFRL, AFOSR invites proposals for research in the areas described in detail below.

The schedule for this announcement is given in Section B, Federal Award Information. This research effort will consist of multidisciplinary teams of researchers with the skills needed to address the relevant research challenges necessary to meet the program’s goals. Multi- investigator teaming is encouraged.

Multi-university teams are allowed. Under no circumstances will the Government help to create teams. In DoD missions, the exponential growth of data demands advanced data analysis capabilities with higher processing performance, lower energy dissipation, and better system scalability.

During the past twenty years, the amount of data to be processed has doubled every 3-4 months while performance of the processors has doubled roughly only every 3-4 years (and will 4 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 probably not continue to do so due to the end of Moore’s Law).

In addition, the intrinsic limitations of the von Neumann computing architecture based on CMOS technology are prohibiting our computing platforms from meeting future artificial intelligence (AI), data analytics and autonomy requirements.

These limitations have motivated emerging research areas in neuromorphic computing inspired by the architecture and mechanisms of the human brain, which is widely recognized as the ultimate computing engine with extremely high energy efficiency, reliability, efficient learning and robust cognitive abilities [1][2].

The human brain’s cognitive functions emerge from the collective processing capability of simple computing elements, i.e., neurons, synapses and dendrites. In such biological neural networks, spike sequences carry both spatial and temporal information for communication and processing.

Moreover, neurons operate asynchronously in an event-driven manner, and biological neural networks demonstrate very energy-efficient information processing. Such biological neural networks have initially been simulated via software-only approaches. The scale of simulated networks usually is small due to high communication overhead and limited parallelizable computing with conventional hardware.

The artificial neural network (ANN) approach loosely models neuron functionality and the massive connection of neurons in a biological brain but ignores a lot of essential features of biological neural networks. Such a simplification makes the training process quite subtle, inefficient and sensitive.

Although ANNs have obtained substantial successes in many applications such as image and speech recognition, the incredibly high computing cost required by training and poor support of in-hardware learning emerge as key obstacles [4][5][6][7][8]. Neuromorphic models differ from their ANN counterparts in that they encode information via the temporal activation of neurons and precise emission of spikes.

In recent years, many research efforts have been devoted to neuromorphic models to harness their higher computational potential [12].

These efforts have focused on developing algorithms such as supervised plasticity rules for precise temporal spike-train recognition, backpropagation approximations for powerful data-driven learning, and spike-timing-dependent plasticity (STDP) for scalable temporal learning in progressively deeper neural network architectures.

CMOS-based hardware has been built to implement the neuromorphic models and shown limited architectural design perspectives [7][9][10][12]. The most visible commercial examples are IBM’s TrueNorth neurosynaptic processor [10] and Intel’s Loihi neuromorphic chip [11]. CMOS devices and circuits, as the building elements for these hardware systems, were not or optimized for neuromorphic computing purposes in the first place.

Therefore, the existing approaches are not able to address the following deep scientific and technological gaps: Gap 1: Methods to realize bio-realistic algorithms and models.

The algorithms and models used in biological neural networks (and their co-evolution with the underlying “wetware”) are likely why the brain is so efficient in performing many tasks such as learning, which cannot be efficiently realized with high fidelity using CMOS hardware. For instance, learning has not been an integral part of the existing neuromorphic architectures.

The backpropagation learning in a deep neural network model requires a substantial amount of high-precision computing and memory, which are not affordable, especially for edge and IoT devices with limited resources. CMOS devices function by transporting electrons, which are volatile, and lack native learning and memory capabilities.

As a result, most of the existing architectures only support inferencing 5 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 functions of pre-trained neural networks that lack the adaptability to perform well in real-world environments with dynamic data flows and characteristics.

Biologically plausible algorithms and models need to be further investigated to identify key computing elements and neuromorphic dynamics that can guide novel hardware research and potentially yield novel computing architectures. Such novel hardware with the appropriate equivalent circuits can more faithfully implement neuroscience principles and thus be much more efficient.

Moreover, the new hardware can serve as a testbed to verify neuroscience principles. Compared to the living tissue, which is still more or less a “black box,” such bio-realistic hardware is a “white box,” where every node of the network can be monitored and modeled for understanding the collective behavior of the system. This will in turn accelerate biological model development.

Such bio- realistic neural networks will not only benefit from but also improve the understanding of the brain, leading to a virtuous cycle for neuromorphic computing and neuroscience discoveries to enhance and accelerate each other. Gap 2: Next generation bio-inspired materials with neuromorphic dynamics.

Bio-inspired materials that can fundamentally enable bio-realistic algorithms and models are still under- researched despite progress with metal oxides, polymer composites, photonics and biomolecules under the DARPA SyNAPSE program and recent AFOSR Multidisciplinary University Research Initiative (MURI) grants ( Brain-Inspired Networks for Multifunctional Intelligent Systems and Cross-disciplinary Electronic-ionic Research Enabling Biologically Realistic Autonomous Learning ).

When relating these materials to CMOS devices, it is important to recognize that CMOS technologies were created for logic operation and arithmetic computation, and thus their dynamics are not well matched with the dynamics that are critical for neuromorphic computing. The dynamics of neurons, synapses and dendrites come from ion diffusion, which is not present in CMOS devices.

Consequently, CMOS-based synapses and neurons require complex and bulky circuits built with transistors. Compared with the 10 μm 2 neuron area, 0. 001 μm 2 synaptic area and ~2 fJ synaptic operation energy in biological systems, the CMOS-based elements are 20 times, 400 times and 2,000 times larger, respectively [11].

Novel intelligent materials, such as memristive materials providing ion diffusion dynamics, could offer such desirable dynamics to implement bio-inspired algorithms and models Neuromorphic photonics (application of photonic principles to the neuromorphic domain) has recently emerged as another possible solution to the shortcomings of state-of-the-art neuromorphic architectures [68].

This new field combines the advantages of photonics and neuromorphic architectures to build systems with high efficiency, high interconnectivity, and high information density and paves the way to ultrafast, power-efficient and low-cost, and complex signal processing. There is a need for research into a photonic memristor towards the goal of realizing the next generation of full-optical neuromorphic hardware [69].

However, the fabrication of custom designs with nanometric features for achieving functional neuromorphic units remains challenging, especially at the biological synapse scale. Achieving laser patterned photonic memristors with sizes comparable to those of the biological synapses (20–40 nm) will require fabrication methods that would go beyond the diffraction limit [70].

Technologies such as super-resolution photoinduction-inhibited nanofabrication (SPIN) potentially would enable rapid prototyping of two-dimensional (2D) and three-dimensional (3D) structures with a resolution well below the diffraction limit. A two-beam nanolithography technique can also be 6 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 used to achieve super-resolution patterns [71].

Gap 3: Innovative circuits to serve as required computing elements. The bio-inspired intelligent materials need to be built into new electrical and/or photonic circuits as required computing elements that can implement the bio-realistic algorithms and models.

Different from the traditional linear circuits built with CMOS devices, bio-realistic circuit designs should be highly nonlinear and rich in dynamics, which present great challenges in unconventional circuit designs. There is a lack of research to take advantage of the intrinsic nonlinear dynamics in designing novel circuit elements to efficiently and faithfully emulate synapses, neurons, and Gap 4: Scalable computing architecture.

Even with more capable and suitable devices and circuits, there is a lack of research and existing knowledge on how to construct a scalable architecture for real-world problems. It requires co-designs and co-optimizations across different system abstract layers to incorporate the desired properties of the new materials and circuits to realize an efficient, reliable, and scalable computing system. Take reliability as an example.

Most of the existing neuromorphic architectures adopt the traditional design method with deterministic data representation and operations. However, the stochastic nature of spike patterns is an important dynamical feature of neuron functions.

A spiking-based model without stochastic elements will have not only limited performance potential in probabilistic inference- related applications but also lower resiliency to noise and achieve less robust performance in real-world applications.

The neuromorphic dynamics of the new materials would be intrinsically stochastic and could serve as the stochastic elements to solve this issue if appropriately incorporated into the new architecture designs.

This COE will investigate, discover and design revolutionary 1) biologically realistic algorithms/models, 2) enabling intelligent materials and devices, 3) more compact but capable computing elements, and 4) scalable and reliable architectures to overcome the science and technology gaps mentioned above. It shall also foster close relationships between AFRL and faculty members as well as their graduate students from top universities. 3.

OBJECTIVE AND RESEARCH CONCENTRATION AREAS This COE aims to support high-risk, high-reward basic research that will address the hardest challenges currently facing neuromorphic (brain-inspired) computing.

Specifically, the basic research objectives of this COE include: (1) Explore and understand bio-realistic algorithms and models to identify key computing elements and neuromorphic dynamics; (2) Discover intelligent materials and devices with intrinsic dynamics to enable bio-realistic algorithms and models; (3) by algorithms and models; and (4) Explore and design scalable, reliable architectures for bio- realistic algorithms and models.

Proposals must address all four research objectives to be considered eligible for funding. Research Objective 1 : It is well established that neurons communicate information via “spikes” (or action potentials).

Spiking neural networks (SNNs) are increasingly common algorithms that use simulated spikes to encode and communicate information and attempt to mimic the signals 7 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 found in biological brains. The use of spikes is not only biologically-inspired but also improves the overall efficiency of the models.

However, debates continue among neuroscientists regarding whether neurons use precise spike timing or frequency to encode information, the functions of noisy, probabilistic population codes, and whether every spike carries information. Effective information representation is expected to correlate with the neuromorphic dynamics and the corresponding circuit scheme based on intelligent materials.

Moreover, many ANN design principles were inspired by the human brain, such as long short-term memory (LSTM) networks [59] and attention models [60][62][63]. However, redesigning and optimizing the cognitive algorithms for realizing these mechanisms in bio-inspired hardware is not straightforward, and a biologically-plausible substitute for backpropagation learning is needed.

Such a substitute would ideally be a multi-layer algorithm capable of training SNNs constructed of various types of connections and data representations that supports both supervised and unsupervised learning. It is desirable that the algorithm should be not only biologically realistic but also efficient to realize in hardware.

An efficient learning rule is likely to take advantage of the dynamics of the neurons and synaptic circuits implemented with new materials. Bio-inspired systems are poised to excel in continuous time, cognition-with-context tasks such as sensory information processing and navigation in real-world environments.

An ultimate objective is to create information processing and learning methods that can adapt to time-varying contexts and environments by leveraging spatiotemporal information. Research Objective 2 : Existing research and demonstrations of computing with memristive materials and devices are mostly limited to using their steady-state (static) behaviors for learning and inference.

However, dynamics of ion motion in a biological neural network play a pivotal role in implementing neuroscience principles for the brain, which represents the most efficient and intelligent computing system ever known.

Such dynamic properties can naturally encode temporal information and are highly desirable for implementing bio-realistic neural networks for applications such as time sequence prediction and natural language understanding. As with the corresponding biological components, memristive materials function based on particle motion.

This makes it possible for memristive materials to generate dynamics like those in biological systems, leading to materials capable of implementing advanced learning and memory functions. The emphasis of biological ion dynamics here requires new memristive materials to be different from the existing ones used for static, non-volatile memories.

More specifically, both the mobile species and materials that host the mobile species need to be discovered for memristive materials with tunable activation energies and a variety of particle motion dynamics.

Different from previous studies for steady-state properties, where only the static resistance states are relevant, the entire temporal switching process, i.e., the switching dynamics, should be utilized for computing and needs to be fully investigated and modeled. This imposes much greater challenges in characterizing, understanding, and modeling the memristive materials.

For example, novel in-situ material characterization methods with very high temporal and spatial resolutions may be needed to reveal the microscopic picture of element migrations. Photonic approaches such as those mentioned above may also be relevant. These experimental observations can then help to build models for mechanism and process understanding.

Such understanding will provide criteria and guidance for new material investigations. Research Objective 3 : The rich dynamics of memristive materials will enable different types of neuromorphic devices and circuits. For instance, the delay time and relaxation time of a volatile 8 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 memristive device are important to realize a faithful artificial neuron and synapse, respectively.

Concepts from the random recurrent neural network (RNN), optical or photonic computing, and reservoir computing communities may offer alternative paths to exploit the inherent nonlinear dynamics of neuromorphic circuitry and devices. As an example, reservoir computing is a machine learning method based on RNNs that use topological diversity and nonlinearity in dynamical processes to perform computational mappings [64][65][66].

Instead of tuning the weights throughout the entire RNN to achieve a target mapping (an expensive and numerically unstable process), only the readout layer of the network is trained using inexpensive linear regression methods. In addition, the diversity requirement of the temporal network properties may inherently exploit variations in fabrication and operation of experimental systems.

These advantages open the door to embedding RNN-like functionality in unconventional physics and hardware [67], allowing a neuromorphic circuit to directly serve as the sensor platform and data processor. Finally, in contrast to CMOS based architecture, memristive dynamical devices can be readily stacked to realize 3D circuit designs, which are necessary for a complex computing system with massive connectivity and computing.

Research Objective 4 : A biological neural network is a complex system with rich dynamics that maps signals in time and space. The efficiency of such networks can be attributed to the biological mechanisms of the neurons, dendrites, and synapses dedicated to these tasks. Realizing the same dynamics in an electronic circuit is more complicated.

As mentioned before, current silicon implementations of neurosynaptic dynamics involve intricate circuit topologies composed of multiple devices. However, these circuits require a large chip area and consume relatively high power, and their performance is more susceptible to noise and environmental variations.

We seek circuit and architecture solutions that have a low complexity but can faithfully emulate the complex dynamics involved in biologically-plausible neurosynaptic computation and learning [47][48]. Such hardware shall take advantage of novel neuronal, dendritic and synaptic circuits that are able to capture and store temporal information as well as modulate spiking voltage inputs in a predictable way.

For example, an architecture can mimic the behavior in a biological circuit, such as in some models where the timing relationships of pre-synaptic and post-synaptic spikes alter the weight of a synapse and modulate the effect of the pre-synaptic neuron on the post-synaptic neuron.

Auxiliary circuits and architectures would be necessary to leverage these devices properly, enable system scaling, and virtually integrate with algorithms and applications [47][51].

For example, given a network of novel neuron and synaptic circuits, a recurrent spiking neural network architecture would require auxiliary circuits for denoising, encoding, and decoding spiking data, a methodology for storing spike times, and a pipeline or schedule for transferring the data between layers.

How to incorporate innovative circuits that leverage novel materials to create revolutionary brain-derived computing architectures would be our ultimate interest. The input/output interface and networks primarily optimized for enhancing the transition of spiking signals would be necessary, too. Moreover, the reliability and robustness of such systems emerges as a major concern [52].

At the system level, predictable, replicable, and robust function and performance are expected. Close dialogue and co-design of materials, devices, circuits, architectures, and algorithms are important for achieving this goal. 9 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 4.

UNIVERSITY CENTER OF EXCELLENCE Proposals for this COE are sought that articulate the technical details of the proposed research and the planned mechanisms to educate a new generation of professionals. Described below is what a strong proposal should provide. 4.

1 Technical Details of Proposed Research A strong proposal should outline, as specifically as possible, the technical details of the proposed research. In addition, proposals should articulate how research goals align with the four research objectives outlined in the previous section.

Proposers can enhance or deviate from these topics if they provide reasonable technical arguments and if they are still addressing the required objectives. Efforts integrating a variety of approaches are preferred to those using a single approach, especially as these may provide increased opportunities for information and technology transfer to AFRL. 4.

2 Investigator Qualifications In line with the basic research vision of this COE, proposals should highlight the following qualifications of the proposed academic collaborators: 1. Strong history of published research that is both principled and foundational in the fields of bio-inspired models and algorithms, neuromorphic computing architectures, and/or dynamic nanomaterials, nonlinear computing circuits. 2.

Expertise within the academic team covering the range of computer science, neuroscience, materials science and electrical engineering with potential contributions from mathematics 3. Strong history of applying principled neuromorphic models, self-learning algorithms, dynamic data analytics, and/or hardware architectures for learning to real-world problem 4.

Willingness to collaborate on and commit resources to jointly defined projects with AFRL scientists and engineers (S&Es). 4. 3 Interactions and Information Exchange Proposals should address plans for the following interactions between AFRL S&Es and 1.

One or more research projects, jointly defined by the academic collaborators and S&Es at AFRL that are in line with collaborators’ expertise, AFRL interest, and the goals of this COE. 2. Commitment from academic collaborators of graduate students and/or post-doctoral associates to work on the aforementioned project(s) both at the academic institutions and embedded with S&Es at AFRL, working on USAF and USSF data and learning tasks.

3. A yearly workshop, independent of annual meetings, where topics are jointly defined by the academic collaborators and AFRL S&Es. Such a workshop would be organized by the academic collaborators, include invited speakers, and be located at an AF/DoD facility.

4. If applicable, how the COE will leverage other institutional resources to expand the participation of students and postdoctoral fellows in the COE, establish dedicated facility and office space, and/or other means of promoting research. Proposers are encouraged to confer with the designated points of contact as soon as possible.

Their contact information can be found at the end of this announcement. Coordination with 10 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 AFOSR, AFRL/RI, and AFRL/RX prior to proposal submission is encouraged but is not 5. ACCESS TO DOD RESOURCES Proposals may request access to AFRL facilities or DOD high performance computing resources in order to conduct the proposed research.

Proposals should make this request in accordance with the instructions given in the D. 4. g.

Project Narrative section of this announcement. If authorized, there is no cost to the research for these resources. Applicants are advised that routine access of educational institution researchers to AFRL/RI and AFRL/RX buildings and facilities is limited to U.S. citizens.

Individuals eligible for access are subject to background checks. Section C. 3.

a. Research Personnel Facility Access Requirements and Restrictions provides more information. Award(s) under this FOA are not restricted in the use of US and non-US citizens, but access to DoD facilities is limited for non-US citizens, which could add coordination challenges.

An objective of this COE is to establish relationships between researchers at the performing universities and the relevant AFRL Directorates. If relevant, proposals may include information on US and non-US personnel, describing their roles in the research effort. B.

FEDERAL AWARD INFORMATION AFOSR anticipates making one award under this announcement, through an issued grant. Any award made under this competition will support a University Center of Excellence in Brain- Derived Neuromorphic Computing with Intelligent Materials and is subject to availability of funds. AFOSR executes discretionary research and development funds appropriated to the USAF and the USSF for awards.

AFOSR can only make an award if sufficient funds are The anticipated period of performance is a three-year base period, with one two-year option to continue performance. As a result, the total period of performance if all options are exercised is AFOSR anticipates not more than $1,000,000 per year in funding may be made available to fund one (1) award from the proposals received.

This plan means proposers should plan on not more than $5,000,000 in funding for the entire five-year duration if all options are exercised; however, the total amount of funding and resources made available to fund a successful proposal may vary based on the quality of proposals received, and funds availability. AFOSR reserves the right to select and fund for award all, some, part, or none of the proposals received.

There is no guarantee of an award. Our authority for an award under this competition is established at 10 U.S.C. 2192(b)(1)(B) for improvement of education in technical fields, and 10 U.S.C.

2358 for basic and applied research. We discuss regulations, terms, and conditions that generally apply to our awards in Section F. Federal Award Administration Information.

11 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 C. ELIGIBILITY INFORMATION a. Qualified and Responsible United States Educational Institutions You are eligible to submit an application if you are a qualified and responsible educational institution in the United States as defined at 10 U.S.C.

2194, or an entity comprised of such educational institutions. Educational institution means a local educational agency, college, university, or any other nonprofit institution dedicated to improving science, mathematics, and engineering education. No other entities are eligible to submit applications under this competition.

Any entities receiving subawards must meet these same criteria. Since the intent of this FOA is to fund a research center with co-located researchers and facilities, at least 40 percent of key personnel (i.e., PIs and co-PIs) for a proposed COE must be primarily employed by the institution submitting the proposal.

AFOSR reviews your application, proposal, and Office of Management and Budget (OMB) designated repositories of government-wide public and non-public data, including comments you have made, as required by 31 U.S.C. 3321 and 41 U.S.C. 2313 and described in 2 CFR 200.

205 and 32 CFR 22. 410 to assess risk posed by applicants, and confirm applicants are qualified, responsible, and eligible to receive an award. b.

HBCU/MI, Tribal College and University Applicants Encouraged Historically Black Colleges and Universities and Minority Serving Institutions (HBCUs/MSIs) and Tribal Colleges and Universities are encouraged to submit research proposals and join others in submitting proposals. However, no funds under this announcement are reserved or otherwise set-aside for any specific entity type. The Air Force will only use the E.

1. Criteria for None of the following entity types are eligible to submit proposals as primary award recipients (1) Federally Funded Research and Development Centers (FFRDCs) (2) Individual persons or people (3) Federal agencies (to include Military Educational Institutions) 2. COST SHARING OR MATCHING Cost sharing or matching is neither required nor an evaluation criterion for proposals under this announcement.

Leveraging other institutional resources to increase the participation of students and postdoctoral fellows in the COE and/or promote research and relationships with AFRL to benefit the COE would enhance the collaboration plan that will be evaluated as part of the DoD relevance criterion for proposals under this announcement. 12 FOA-AFRL-AFOSR-2021-0005 Amendment 0001 a.

Research Personnel Facility Access Requirements and Restrictions AFRL contains facilities and equipment that could be useful to this Center. Access to these facilities will be restricted to US citizens or permanent residents. b.

Acknowledgment of Support and Disclaimer Requirements You must include the F. 3. d.

Acknowledgment of Research Support on all materials created or produced under our awards. The F. 3.

e. Disclaimer Language must be included on materials as required. The award document may provide additional instructions about specific distribution statements to use when you provide research materials to us.

You are not eligible to submit a proposal