HR001124S0022_ScAN_FAQ_Updated_06_21_2024.pdf

Scalable Analog Neural-networks (ScAN) Program

Frequently Asked Ques�ons Document

Updated on June 21, 2024

1. Ques�on: Does a subcontractor require a CAGE or NCAGE code? Is the subcontractor also obligated to report finances and undergo audits related to the program funding?

Answer: CAGE codes are required with a few excep�ons for subcontractors working on unclassified efforts. The informa�on requirements will change depending on the contrac�ng instrument selected for award by the prime in these cases. Please review DARPA Connect for addi�onal informa�on.

2. Ques�on: Can a commercial company be a sub-contractor?

Answer: An organiza�on/company/individual can be on mul�ple proposals, including a commercial company.

3. Ques�ons: [Sensor Development]

• Are you looking for an end-to-end processing system that directly interface to an analog sensor?

• Do we have to design an image pixel array to go along with the analog processor?

• Will design of an image sensor such as specialized ROIC within the program scope?

Answer: ScAN will neither develop an analog sensor nor an interface to the sensor, such as a specialized ROIC. Rather, analog signals from the sensor will be simulated. Please see the Broad Agency Announcement (BAA) for more details on program-specific requirements.

4. Ques�ons: [Device Technology]

• What technologies are considered for the program, conven�onal CMOS or any new device technologies?

• Are we allowed to use a CMOS technology that offers a device type that may not be available in all CMOS technology nodes?

• Is In-pixel compu�ng using non-CMOS, e.g. compound semiconductor and heterogenous packaging with CMOS chip, within the program scope?

Answer: The ScAN program does not prescribe technology choices. However, the proposer must explain how the technology will meet the program objec�ves. Note that certain technologies and device technology development are out of scope - as described in the BAA.

5. Ques�ons: [Foundry Service]

• Are fabrica�on and demonstra�on required? If so, at what scale, what technology? Will

DARPA provide any assistance in connec�ng performers with a foundry?

• Are there any limita�ons to the process node or foundry selec�on?

Answer: Fabrica�on and demonstra�on are required. The scales of demonstra�on are defined in the BAA. There is no preferred foundry selec�on or process node. The ScAN program does not plan to provide specific Government Furnished Capabili�es (GFC), such as foundry service assistance.

6. Ques�ons: [Target Applica�ons and Datasets]

• Are there any primary targeted applica�ons, algorithms, or use scenarios?

• Should we expect DARPA to provide relevant data sets to performers as inputs to ScAN or will the performers be providing their own sensor data. If the later, is there any preference for specific spectral bands or other sensor characteris�cs?

• What type of demonstra�on are sought for the so�ware and hardware (benchtop) at each phase? Will data sets be provided?

• Will there be a test use case defined at each phase?

• You have only described image processing in the applica�on space, will IR processing be a use case as well? What about others (LIDAR, mul�-spectral cameras, etc.)?

• Could the proposal be writen to address a specific applica�on, such as image classifica�on or object detec�on and tracking?

• Is there a pre-defined list of neural networks for Phase 1 and Phase 2 that need to be supported? What degree of flexibility/programmability needs to be supported in the neural network implementa�on?

• Do we have to use a par�cular architecture of the neural network? Is it convolu�onal net?

• Will there be interest in novel AI algorithms, such as neuro-symbolic or hyperdimensional compu�ng, where the algorithms can demonstrate natural robustness to noise in analog hardware?

• There was a lot of discussion about image processing. Are proposals focusing on RFML (ML applied to wideband microwave/RF signals), i.e., applica�ons to RF sensors instead of image sensors in scope?

• How are test examples going to be delivered to the hardware?

• Can you comment on the type of sensing modali�es that are of interest? Most of the examples listed were image or vision based. Are other sensing modali�es and applica�ons spaces of interest, such as audio, radar, etc.?

Answer: There is no preferred architecture or algorithm. The proposed approaches are required to meet the program objec�ves as defined in the BAA. For the purposes of test and evalua�on, the BAA specifies benchmarking datasets for each program phase. The Government will provide the benchmarking datasets as Government Furnished Informa�on (GFI) to emulate analog sensors. The proposed hardware demonstra�on will be required to accept electrical current from those nominal sensors as input. Please see the BAA for more details on program-specific requirements.

7. Ques�ons: [Security Clearance]

• Will performers need to have any security clearance or any par�cular secure facili�es?

• When does DARPA expect to release the ScAN BAA and will there be a classified addendum?

Answer: The ScAN program is unclassified and does not expect any classified effort will be required to meet program objec�ves. Thus, neither security clearances nor secure facili�es are expected to be required. The BAA, as released on 22 May 2024, does not include a classified addendum.

8. Ques�on: Would you be open to proposals for Neural Network architecture research that target mixed analog + digital systems?

Answer: The ScAN program seeks development of all analog and hybrid analog approaches that will achieve the ScAN program objec�ves (Metrics and Milestones), and that are not specifically excluded per the BAA. Note that hardware demonstra�ons are required.

9. Ques�on: For each proposal, do you have a rough distribu�on about so�ware and hardware %? Or it is open for any combina�on?

Answer: There is no preferred distribu�on of effort between hardware and so�ware.

10. Ques�ons: [Photonics]

• What role does DARPA expect photonics to play in this BAA?

• Would scalable integrated photonic architectures be considered?

• Is ScAN interested in only analog circuits or are passive op�cal components that perform neural network ac�vi�es of interest?

• Are ac�ve op�cal approaches coupled with analog backends of interest?

Answer: There is no preferred technology, but certain technologies and device technology development are out of scope as defined by the BAA. While DARPA does not prescribe technology choices, the Government will provide the benchmarking datasets as GFI to emulate analog sensors. The proposed hardware demonstra�on will be required to accept electrical current from those nominal sensors as input. Please see the BAA for more details on program-specific requirements.

11. Ques�ons: [Non-Vola�le Memory]

• Several technologies are listed as ‘out of scope’, including non-vola�le memory-based compu�ng. Neural networks do need memory, vola�le or non-vola�le, to store weights, etc. Could you elaborate a bit more on this?

• Can you clarify what you mean by “Non-vola�le memory-based compu�ng” is out of scope? Do you narrow down to conven�onal MAC with nonvola�le memory or any design that involves nonvola�le memory?

• Are you only interested in SRAM and DRAM based approaches?

• Are all non-vola�le memories excluded or is this statement targeted specifically for passive (resis�ve) non-vola�le memories?

• Does the BAA exclude approaches where weights are stored in the compu�ng devices?

Answer: Teams may propose the use of any memory technology to store weights that will achieve the program objec�ves (Metrics and Milestones). Non-vola�le memory (NVM), when used for mul�ply-accumulate (MAC) or similar func�ons, is out of scope as defined in the BAA.

12. Ques�on: Can there be an early tape-out, say to a standard CMOS fabricator, in Phase 1a to help decrease the risk of fabrica�on in Phase 1b and Phase 2?

Answer: Given limited resources and compressed schedule, cri�cal design reviews are required prior to any tape-out as defined by the BAA. Please see the BAA for more details.

13. Ques�ons: [Input Interface]

• Does the program specifically target an electronic chip that processes electrical input and output? Would a system with op�cal input and electrical output, or electrical input and op�cal output, also fall within the program's scope?

• Is development of any op�cal analog pre-processing subsystems in scope, or do you only want innova�ons in electronics, not op�cs?

• Will the demonstra�ons in Phase 1 and Phase 2 be performed using real-�me sensor outputs (for example, by integra�ng with an image sensor) or using “digital inputs stored in a memory”?

Answer: The benchmarking datasets will be provided as GFI to emulate the analog sensor for the purposes of tes�ng and evalua�on. The proposed hardware demonstra�on will be required to accept electrical current as input from those nominal sensors. Please see the BAA for more details on program-specific requirements.

14. Ques�ons: [Opera�on Range]

• Does the program have size or reliability targets (such as vibra�on resistance and opera�ng temperature range) for "ac�onable intelligence in the field"?

• What is the targeted temperature range for opera�on? Is it the same as for the conven�onal CMOS (-55 C to 125 C)

Answer: The ScAN program does not specify physical size or reliability targets such as vibra�on resistance. While the ScAN BAA does not specify an opera�ng temperature range, the ScAN program will ini�ally target -20°C ~ +85°C, and further research the applicable range during the effort. Please see the BAA for more informa�on on metrics and milestones.

15. Ques�on: I believe I caught that solu�ons that would imply new technology templates for foundry processes are discouraged. As in, there’s no room, especially in Phase I, for solu�ons that try to forge new processes that don’t already exist at available foundries. With that in mind, would solu�ons that u�lize ac�ve circuits be welcome if they conform to *exis�ng* ASIC processes and pla�orms?

Answer: Correct, however, there is no preferred circuit approach or technology pla�orm.

16. Ques�on: Is the applica�on for ScAN video image classifica�on/inference? There are some NN architectures that are extremely good for such applica�ons (for instance in-pixel NN camera architectures) but not necessarily for general purpose NN architectures.

Answer: Please see the BAA for more informa�on on benchmarking datasets.

17. Ques�on: If a proposed solu�on is more advantageous for >> 100 M parameters but not as advantageous for < 20M parameters, would it be within scope for DARPA ScAN?

Answer: Please see the BAA for more informa�on on model size requirements.

18. Ques�on: Does [NIST’s] superconduc�ng solu�on require the typical ultra-low temp opera�on of the essen�al circuit components? If so, there goes the edge-compu�ng poten�al.

Answer: As stated in the ScAN program Special No�ce, DARPA-SN-24-57, neither DARPA nor the DoD endorses the informa�on, or the organiza�ons contained in the virtual presenta�on, nor does DARPA or the DoD exercise any responsibility for improper dissemina�on of the virtual presenta�on.

19. Ques�on: How important is it to transi�on to a program or have a customer demonstra�on?

Answer: Program transi�on and demonstra�on is an integral part of the evalua�on criteria under sec�on “Poten�al Contribu�on and Relevance to the DARPA Mission.” Please see the BAA for more informa�on.

20. Ques�on: Accuracy suggests a classifica�on task. Is localiza�on of objects in the input not required?

Answer: Please see the BAA for more details on program-specified requirements.

21. Ques�ons: [Budget]

• Did we hear correctly that the technology process (foundry and technology node) that we target is up to the proposers and all fabrica�on costs should be included in the proposal?

• Should all IP costs (eg, DDR controllers) be included in the proposal budget?

• Should the costs of a board level demonstra�on be included in the budget?

Answer: All costs that are associated or needed to achieve the BAA milestones, including demonstra�ons, should be included in the proposed budget. The ScAN program does not prescribe technology choices and does not plan to provide foundry service assistance as Government Furnished Capabili�es (GFC).

22. Ques�on: During the presenta�on, it was men�oned to consider classifica�on problem in Phase I and extend to object detec�on and tracking problem in Phase II. Are there any suggested datasets to demonstrate the proposed capabili�es? In the ScAN program, are there any suggested datasets for performance evalua�on? This problem �es to the MNIST dataset point men�oned during the talk as well, where we want a sufficiently challenging dataset that goes beyond toy problems.

Answer: Please see the BAA for more details on benchmarking datasets and program-specific requirements.

23. Ques�on: In addi�on to PVT varia�ons, ADC offset is another challenge in analog inferencing. Are solu�ons that address that challenge considered within the program scope?

Answer: The ScAN program does not prescribe a specific circuit approach, but the proposer must describe how their chosen approach will meet the program objec�ves (Metrics and Milestones). Please see the BAA for more details.

Addi�ons on June 21, 2024

24. Ques�on: Are electro-photonic solu�ons acceptable for the ScAN program or is the ScAN program exclusively looking for purely electronic solu�ons?

Answer: The ScAN program does not prescribe technology choices except for those listed as out-of-scope technologies and device technology development efforts. However, the proposer must explain how the chosen technology, whatever it may be, will meet the program objec�ves.

25. Ques�on: Is there any upper limit on the budget for the ScAN program?

Answer: There is no upper limit or cap on proposed budgets. However, please note, that all proposals will be evaluated based on “Cost Realism” as stated in Sec�on II of the ScAN BAA.

26. Ques�on: When/where an FAQ will be available?

Answer: The FAQ is published under the “Atachments/Links” sec�on of the ScAN BAA page on SAM.gov. A direct link to the FAQ on SAM.gov is posted under the ScAN announcement on the DARPA Opportuni�es website. Updates to the FAQ will be annotated as such and be posted periodically to these same loca�ons.

SAM.gov: htps://sam.gov/opp/c6687528fcc54113a33256ec445cd0a2/view

DARPA Opportuni�es: htps://www.darpa.mil/work-with-us/opportuni�es?oFilter=MTO https://sam.gov/opp/c6687528fcc54113a33256ec445cd0a2/view https://www.darpa.mil/work-with-us/opportunities?oFilter=MTO

27. Ques�on: [Proposers’ Day Materials]

• Can you please provide the ScAN Industry Day Slides? They are not on SAM.

• Is there are video replay of the proposer day of ScAN?

Answer: The slides and recording from the ScAN program Proposers’ Day are posted under the ScAN BAA announcement on SAM.gov and the DARPA Opportuni�es website.

DARPA Opportuni�es: htps://www.darpa.mil/work-with-us/opportuni�es?oFilter=MTO

SAM.gov: htps://sam.gov/opp/c6687528fcc54113a33256ec445cd0a2/view

28. Ques�on: The BAA indicates that "diverse exper�se from across academia, small business, and the defense industrial base research community may be applicable to achieving all program goals." Does it have to be interpreted into the fact that presen�ng a proposal that involves a large alpha customer/early adopter part of the Defense industrial base would be a key criterion for DARPA?

Answer: The ScAN program BAA does not require par�cipa�on/inclusion of customers or early adopters. Rather, the proposed teams’ exper�se may be sourced from any combina�on of academia, small business, and the defense industry-based research community so as to form a diverse and comprehensive team to achieve all program goals.

29. Ques�on: We have a few ques�ons that we would like to have clarifica�on on:

• Phase 1 benchmark target: In Sec�on 1.1.3.1 and Table 1, it is men�oned that performers will develop a demonstra�on system capable of handling 3,000 inputs, with the analog NN demonstra�on system needing to be equivalent to a digital model of 1M parameters. Our specific ques�on is: What does the "90% accuracy of the state of the art" requirement in Table 2 refer to? Is "state of the art" referring to a model of at least the size specified in Table 1 and similar architecture? e.g. If different models have varying levels of accuracy and parameter counts (>1M), how should we determine which model to use for comparison (accuracy vs number of parameters)? What if the model's accuracy outperforms other models but with significantly different architectures?

• Phase 2 benchmark target and technical challenge: In Sec�on 1.1.3 and Table 2, the Phase 2 benchmark target includes working with a BDD image dataset for object detec�on and tracking. The BDD100K dataset provides 10 tasks. Could you please clarify which specific tasks are required with the BDD100K dataset? Specifically, are there detailed requirements or expected outcomes for the object detec�on and tracking tasks that need to be addressed? We want to ensure we understand the nature of the work to be carried out with this dataset.

Answer:

https://www.darpa.mil/work-with-us/opportunities?oFilter=MTO https://sam.gov/opp/c6687528fcc54113a33256ec445cd0a2/view

• Phase 1: There is no preferred or prescribed SoA reference model. It is up to the performer to propose an appropriate model for implementa�on that will best demonstrate the architecture robust to varia�ons and power efficient to meet the program metrics and milestones, and also considers scalability toward Phase 2 target.

• Phase 2: The ScAN program plans to u�lize the BDD100k MOT set as a baseline and will ini�ally target Mul�ple Object Detec�on and Tracking, or similar tasks, for benchmarking.

30. Ques�on: My team has been in conversa�on with collaborators to develop a proposal for the ScAN program. We have a system concept and architecture that can meet the program goals with the possible excep�on of the total system power. Our approach leverages superconduc�ng electronics, so a few hundred wats are required to cool the chip to 4K. We were deba�ng whether your requirement for 1W system power includes cooling power.

Answer: Yes; total system power does include cooling power. The “total system power includes all power consump�on from the neural network and peripheral circuits, and pre-and post-processing circuits (analog and digital).”

31. Ques�on: Are we allowed to use a CMOS technology that offers a device type that may not be available in all CMOS technologies? We note that this device type is a standard foundry offering that is part of this CMOS technology node. It is a device type, however, that may not be present in all CMOS nodes. Is targe�ng such a technology acceptable?

Answer: The ScAN program does not prescribe technology choices except for those listed as out-of-scope technologies and device technology development efforts.

32. Ques�on: We plan to respond to the call and submit an abstract. Could you please introduce how to submit the abstract? Is there a website to submit it online?

Answer: Both abstracts and proposals must be submited using DARPA’s Broad Agency Announcement Tool (BAAT). This can be accessed via the Proposer Instruc�ons: General Terms and Condi�ons (darpa.mil) page, in the Unclassified Submission Instruc�ons sec�on.

https://www.darpa.mil/work-with-us/proposer-instructions https://www.darpa.mil/work-with-us/proposer-instructions