Paper Title: Main Memory: Its Future Direction in the Hybrid Cloud and in AI

Paper Abstract: As AI applications are rapidly increasing, the amount of data processing in computing systems is growing rapidly. The flexibility and capability of processing big data has transformed hybrid cloud platforms into essential building blocks for the IT infrastructure. To achieve high performance, main memory must provide high bandwidth and low latency. DDIMM meets such requirements by having the DRAM controller connect to host processors using the OMI. To process the AI-centric workloads, designing a platform architecture that is flexible and scalable is required at a lower cost. CXL can meet this demand by allowing processors to connect to accelerators and multiple tiers of memory. Enabling low cost memory in the CXL platform is viewed as a means to improve the future systems value proposition. While DRAM scaling has been successful in enabling large memory capacity at a reduced cost, continued scaling has resulted in Row-hammer mechanisms. These scaling issues affect the reliability and security of the infrastructure. The challenges of the hybrid cloud computing are reviewed in this talk through the lens of memory technology, and the direction of future memory technology.

Paper Author: Ju Jin An, STSM, IBM

Author Bio: Ju Jin An is a Senior Technical Staff Member in IBM's Systems Supply Chain Organization. Ju Jin has a background in silicon fabrication process and process integration through 20 years of career in the semiconductor industry. She is responsible for bringing up DRAM memory module technology for IBM power and z system. She holds a MS/Ph.D. degree in Chemical Engineering from MIT and a BS from Seoul National University.