Frank Ferro, Senior Director Product Management at Rambus, and Shane Rau, Senior Research Executive at IDC, recently hosted a webinar that explores the role of tailored DRAM solutions in advancing artificial intelligence. In part one of this four-part series, Ferro and Rau discuss a wide range of topics including the interconnected system landscape, the impact of COVID-19 on the data center, and how AI is helping to make sense of the data deluge.
The Interconnected System Landscape
According to Rau, the internet is “segmenting” with the advent of ubiquitous communications: cellular and Wi-Fi.
“Over the last 10 years, we’ve built ubiquitous communications across different system types with interconnected servers and data centers,” he explains. “For interconnected data centers, we have the cloud connecting through communications infrastructure to PCs, phones, and tablets, but also through edge infrastructure to IoT systems. What we used to call embedded systems are now IoT systems and even cars.”
According to Rau, there is now a landscape of interconnected systems. This means data – and the computing of that data – can be located anywhere across this broad landscape.
“The opportunity for AI – and by extension semiconductors – is vast across the internet system landscape,” he elaborates. “Even when you look at the relatively new segments of the landscape, those would be IoT, and the edge infrastructure systems that bring computing power closer to both IoT and endpoints. The opportunity there is newer, but also growing significantly. And indeed collectively, the revenue for all of those microcontrollers and microprocessors that are running AI – the opportunity at the endpoints and in the edge infrastructure – is more than it is in the traditional data center and cloud.”
COVID-19 & the Data Center
Rau also touched upon IDC’s semiconductor forecast in the context of COVID-19.
“We see in the data center, cloud and infrastructure and edge infrastructure, that there will still be significant investment this year from folks like Google and Amazon in their cloud infrastructure,” he continues. “2020 and 2021 could actually be better years in light of COVID-19 and the need to build out infrastructure and serve a more distributed user base – folks like us who are working from home and edge infrastructure servers – and also servers closer to the endpoints that enable distributed computing and reduce latency.”
The industry, says Rau, clearly sees the need to move data and locate the data closer to the end user. This, he emphasizes, presents an opportunity for more processing silicon.
“Coming out of the crisis, the essential technologies enabling data center and cloud infrastructure and edge infrastructure will include connectivity technologies like 5G,” he states. “A specific example of the need for more processing silicon to support AI capabilities is in healthcare systems for equipment like MRI and other medical imaging systems that need high performance computing to drive real time insights and decision making using large pools of data.”
When we think about AI, said Rau, we should also think about processing those large pools of data.
“Where there is a processor, there needs to be memory and storage. So, we cite memory and storage as one of the major beneficiaries of COVID-19 to move to AI to process data – the processors and the DRAM attached to those processors,” he explains. “More memory means a need for more intelligent memory, and a need for storage and more intelligent storage: real time DRAM serving data processing semiconductors, or storage mechanisms built on NAND such as SSDs and hard drives.”
The Data Deluge and AI
As Rau points out, the ongoing data deluge is the “very basis” for AI. Put simply, there is far too much data for human beings to process on their own with traditional methods.
“By 2024, we’re looking at 143 zettabytes of new data created (in that year). If we further segment the data between data that is original and the copies of that data, 90% of that data will be copies,” he states. “There are organizations, financial and government organizations, that are required by law to have copies of data. As well, we have data center owners who, for reasons of latency, distribute copies of data across the world. By 2024, the ratio of original data to copy data will be about 1:10.”
AI says, Rau, addresses the deluge by intelligently sorting through all the data, helping organizations decide what data to keep, process, and move through the internet system landscape for analysis.
“For example, at an IoT endpoint there is a lot of data created, although not all of that data is going to need to move. When data moves, that costs money, time, and power. AI can be used to sort through that data and decide what is going to move into the data center and cloud for further analysis,” he adds.