2021.09.30 ~ 2022.08.30
09:00am (KST, UTC+9) KR, US
Acceleration DIMM (AXDIMM) is a powerful yet flexible near-memory acceleration platform that allows researchers and developers to implement various functions in memory modules and accelerate memory-intensive applications, such as deep learning recommendation and genomic computation. Samsung is actively seeking out new and innovative applications for AXDIMM by attracting the best talent from universities. We are pleased to announce the official launch of the Samsung Memory Open Innovation Contest.
AXDIMM is a DDR4-compatible, FPGA-based near-memory acceleration platform. A smart buffer chip on a DIMM not only functions as a traditional buffer chip for a high-capacity DIMM, but it also provides ARM CPU cores and FPGA fabrics, which can accelerate various functions. As AXDIMM supports simultaneous access to each rank within a DIMM, it is therefore capable of multiplying the process bandwidth and improving the performance of memory-intensive applications.
Because AXDIMM processes data locally in the DIMM rather than transferring it all the way to the CPU, it also allows for reductions in energy consumption. These qualities make AXDIMM a perfect platform for near-memory processing – one that can easily be integrated with a host through the standard memory channel interface. An example of Samsung's use case is the integration of a deep learning recommend-ation model (DLRM) with AXDIMM, in which the FPGA was programmed to accelerate the embedding operations, proving that AXDIMM can improve performance and energy efficiency.
Reference paper:
“Application-Transparent Near-Memory Processing Architecture with Memory Channel Network”
(51st Annual IEEE/ACM International Symposium on Microarchitecture, 2018)
Samsung has established an ambitious research agenda that highlights two areas of focus which are central to its long-term vision and strategy for AXDIMM.
Samsung seeks to use the AXDIMM architecture to develop as well as recommend hardware and software models for the following applications:
Including recommendation systems, natural language processing, image processing, etc.
Including in-DRAM databases (IMDB) etc.
Five finalists selected on December 15, 2021 will work on their proposed ideas until July 15, 2022. (A system with AXDIMM will be provided virtually during the Contest(R&D) period)
∙ Competition is open to all university professors and students, including part-time / full-time and undergraduate/graduate students in the United States and South Korea
∙ Teams must include a professor from an accredited university.
(One professor can participate on multiple teams.)
∙ Teams have a maximum of 5 participants. (including a supervising professor)
∙ Multiple teams can apply from one university/college.
Proposals should
Be formatted in MS Word (letter size, Times New Roman in 11pt) and include:
Be submitted via email to AXDIMM@samsung.com (Due on Nov 15, 2021)
Infographic that visually shows the prize money of the Open Innovation Contest.
Five finalists will receive a one-time R&D incentive of $10,000
Top 3 teams will receive the following prizes:
1st place: $50,000 / 2nd place: $30,000 / 3rd place: $10,000
Winning ideas may be considered for Samsung research programs after the Contest has concluded
1. Evaluation criteria for initial proposals (5 Finalists)
| Evaluation Criteria | Rate |
|---|---|
| Creativity in the use of AXDIMM in hardware and software architectures | 50% |
| Feasibility of hardware and software architectures, including assessment of the team’s ability to execute the R&D required for their proposal | 50% |
2. Evaluation criteria for Finalists (Top 3 Teams)
| Evaluation Criteria | Rate |
|---|---|
| Suitability of hardware and software architectures for AXDIMM | 30% |
| Innovation of hardware and software architectures | 30% |
| Product feasibility of hardware and software architectures | 40% |
E.g. Memory Channel Computing (MCC)
Enables an ARM processing system within an AXDIMM module
Utilizes Quad ARM Cortex-A53-supported processing system + lightweight OS within AXDIMM
Implements DIMM channel-based network interface – no modification of user application required
Reference paper:
“Application-Transparent Near-Memory Processing Architecture with Memory Channel Network”
(51st Annual IEEE/ACM International Symposium on Microarchitecture, 2018)
E.g. scattering and gathering
Utilizes Quad ARM Cortex-A53-supported processing system + lightweight OS within AXDIMM
Implements DIMM channel-based network interface – no modification of user application required
Reference paper:
“Near-Memory Processing in Action: Accelerating Personalized Recommendation with AXDIMM”
(2021 IEEE_Micro)
Please follow the instructions on the Contest’s webpage, and submit your application to AXDIMM@samsung.com by Nov 15, 2021.
No. The contest is open only to university / college professors and students.
Yes. There is no limit on the number of proposals that can be submitted by one team. Please submit proposals one at a time, and follow the instructions carefully (including signing the terms and conditions) when submitting each one.
Each team may have a maximum of five members, including a supervising professor.
The R&D period begins as soon as the five finalists are selected on December 15, 2021, and ends on July 15, 2022. We encourage teams to adjust the scope of their R&D to fit within the length of the Contest. However, teams may choose to take a phased approach and complete an initial portion of their research and share the interim results at the end of the Contest.
Yes. However, in this case your team will have to choose which university you will represent.
Please reach out to AXDIMM@samsung.com for any administrative or technical questions.
No, the AXDIMM system will not be available during the proposal stage. However, the five teams who are selected to perform R&D will be provided with access to the technology.
Samsung will announce which teams have been selected on the Contest’s webpage, and will also reach out to the selected teams individually.
The AXDIMM system will be provided virtually during the R&D period. Depending on the research topic, we may also consider providing teams with AXDIMM hardware. * The system’s form factor and the duration of access will be determined and communicated by Samsung.
We wish for your continued success in your endeavor to advance
memory system research and inspire technological innovation!
| Subject | Synergistic Approach for Systems with AXDIMMs, GPUs, and NVMe Devices |
|||
|---|---|---|---|---|
| Professor | Jaejin Lee |
|||
| Team members | Heehoon Kim |
Daeyoung Park |
Jinpyo Kim |
Junsik Shin |
| Subject | Near-Memory Acceleration of Layer-5 Network Protocols |
|
|---|---|---|
| Professor | Mohammad Alian |
|
| Team members | Amin Mamandipoor |
Neel Patel |
| Subject | Near Memory CSR Encoding for Sparse Matrix Operation Based on AXDIMM |
|||
|---|---|---|---|---|
| Professor | Ki-Seok Chung |
|||
| Team members | Kwangrae Kim |
Nayeon Kim |
Soomin Rho |
Si-Dong Roh |
We wish for your continued success in your endeavor to advance
memory system research and inspire technological innovation!
Subject |
Synergistic Approach for Systems |
|||
|---|---|---|---|---|
Professor |
Jaejin Lee |
|||
Team members |
Heehoon Kim |
Daeyoung Park |
Jinpyo Kim |
Junsik Shin |
Subject |
Near-Memory Acceleration of Layer-5 Network Protocols |
|
|---|---|---|
Professor |
Mohammad Alian |
|
Team members |
Amin Mamandipoor |
Neel Patel |
Subject |
Near Memory CSR Encoding |
|||
|---|---|---|---|---|
Professor |
Ki-Seok Chung |
|||
Team members |
Kwangrae Kim |
Nayeon Kim |
Soomin Rho |
Si-Dong Roh |
We wish for your continued success in your endeavor to advance
memory system research and inspire technological innovation!
Subject |
Synergistic Approach for Systems with AXDIMMs, GPUs, and NVMe Devices |
|---|---|
Professor |
Jaejin Lee |
Team members |
Heehoon Kim |
Daeyoung Park |
|
Jinpyo Kim |
|
Junsik Shin |
Subject |
Near-Memory Acceleration of Layer-5 Network Protocols |
|---|---|
Professor |
Mohammad Alian |
Team members |
Amin Mamandipoor |
Neel Patel |
Subject |
Near Memory CSR Encoding for Sparse Matrix Operation Based on AXDIMM |
|---|---|
Professor |
Ki-Seok Chung |
Team members |
Kwangrae Kim |
Nayeon Kim |
|
Soomin Rho |
|
Si-Dong Roh |
Global
