Micron Powering the Future of Memory

Article By : Stephen Las Marias

Micron highlights the latest innovations and technology developments shaping the future of memory.

Never has digitalization become more critical than when the global pandemic disrupted the way we work and live. Firms who have embraced digitalization earlier in their journey have been armed with the flexibility, agility, and resiliency needed to minimize the impact of this global issue. Those who were too late in their transformation have now realized how dangerous it is to remain idle in a sea of innovation happening around them. They have learned their lesson and they are now embarking on their digitalization journey and transforming their operations, aided by the proliferation of innovative technologies that are making data more accessible, effective, and useful.

Amid this trend, consulting firm McKinsey & Company projects that advanced data analytics technologies will add $13 trillion to the global economy by 2030—driven by the ability to make better and faster decisions based on the information around us.

Sanjay Mehrotra

“Data is creating newfound opportunities for platform innovation,” said Sanjay Mehrotra, President and CEO of Micron Technology, during his keynote presentation at COMPUTEX 2021. “We have never seen such rapid innovation.”

Mehrotra noted two major trends at the heart of this change: artificial intelligence (AI) and 5G networks. “These two trends interlock—each providing fuel to the other—to create dramatic new potential for data in the post-pandemic world,” he says. “By the end of 2024, 75% of organizations will have shifted from piloting AI solutions to operationalizing them. That shift is driven by the impressive gains AI can provide across all businesses and industries, from breakthrough medical cures to natural language processing.”

Unexpected industries are also using AI to solve real problems, such as in farming, and even in water conservation.

5G, meanwhile, has been one of the most anticipated game-changing trends in technology. While the pandemic hampered 5G network growth in many parts of the world, rollout has resumed, with investment projected to extend even further in 2021. India, for instance, has now started 5G trials in the country.

Over the next five years, operators globally are expected to invest nearly 1 trillion in 5G cellular networks, with individual 5G connections growing to reach 1 billion in 2023 and then 2 billion by the end of 2025, according to GSMA Intelligence.

These two trends are expected to further fuel the explosion of data. And according to Mehrotra, new computing approaches are needed to make best use of the enormous wave of data that we have only begun to create.

Data Center Developments

According to market analyst International Data Corp. (IDC), at least 90% of enterprise applications will embed AI by 2025.

“Whether it’s in conversational interfaces, whether it’s in how they deal with multiple sources of data to train and make predictions versus simply being told to model and simulate, whether it’s in the applications they enable from smart manufacturing to smart ways of dealing with customer relationships and supply chain, artificial intelligence is in everything that these applications do,” says Raj Hazra, Senior Vice President and General Manager, Micron Compute and Networking Business Unit.

This rapid transformation of applications places incredible stress on today’s computing architectures. And the challenge is essentially the exponential growth of the need for infrastructure to keep up with the promise of what AI has to deliver.

Micron is working in partnership with the industry to deliver innovations in memory and storage and ensure the full promise of AI. And this balanced capability between memory and storage is based on two dimensions: scale-up and scale out.

“Scale-up simply means providing more compute, memory, and storage for a single instance of what you want to use the training for,” says Hazra. “It could be a single node in a server or a single shared memory cluster. But as powerful as they can get with advances in memory and storage, bandwidth capacity, and compute performance, they’re still limited with what you can physically, technically, and even economically compute or make available on a single node. The other dimension, while each of these units of AI become more powerful is to connect those units and scale out. This is where hyperscale is really a tremendous move forward.”

Far Memory

Raj Hazra

Algorithms like AI require a much more sophisticated approach to how memory and storage is created for those kinds of workloads. According to Hazra, the near memory that used to be just direct attached is now a combination of in-package memory as well as increasingly more powerful direct attached memory. But there are technical and economical challenge to such approach.

In comes far memory. Targeted by the industry to address the fundamental bottleneck in CPU-centric platform design, far memory is expected to deliver a breakthrough in memory access for data center application requirements.

The creation of the far memory tier is at the heart of a broad platform transformation being brought about by an open industry standard called Compute Express Link or CXL.

According to Hazra, CXL is enabling the ability to add to add both capacity and memory bandwidth, without being constrained to what you can directly attach, and creating new memory tiers, the CXL attached memory tier.

“CXL has gathered broad support due to its capability to provide the architectures critical to run AI workloads at scale, notes Mehrotra. The connectivity CXL provides allows for new degrees of freedom for platform architecture via shared pools of far memory.”

Micron has been collaborating with AMD to help deliver data center innovation. “Modern data center workloads are driving new requirements for infrastructure, and it takes an ecosystem of innovators to work together and solve some of the biggest challenges facing the industry,” says Mark Papermaster, AMD CTO and Executive Vice President for Technology and Engineering. “There’s an insatiable demand for more memory and more compute capability and capacity. The introduction of DDR5 DRAM and advances in platform interfaces like CXL are important milestones to meet customer needs.”

AMD joined the CXL Consortium in 2019 as part of a broader effort to utilize new and emerging memory and storage technologies in an open standards-based solution. “We have a long history of collaboration with Micron, and continued memory technology innovation is key to meeting the demands for compute intensive workloads,” says Papermaster.

Transforming the Edge into a Center of Data

One of the promises of 5G is the enablement of more-capable edge devices. “Computing is moving from data centers to centers of data,” says Mehrotra. “That data-centric shift in capabilities means sharper insights and faster decisions.”

According to IDC, around 73% of enterprises consider the edge a strategic area of investment. “The edge is driving economic opportunity across sectors, expanding services closer to the customer to make more efficient use of apps and data,” Mehrotra adds.

Among services that edge is enabling enterprise to deliver—which they couldn’t before—include more accurate forecasting of retail inventory, faster processing of large healthcare images like MRIs, autonomous platooning of truck convoys, and more immersive cloud gaming.

And both AI and 5G are key factors in fueling this growth and making the edge smarter, enabling its transformation by extending distributed data-centric computing models into new environments, including industrial applications and enhanced automotive experiences.

In fact, apart from providing technology solutions for factory automation, visual analytics, machine-to-machine communication, and even aerospace applications, Micron is likewise using edge technologies in its own semiconductor factories.

For one, the company uses advanced optical inspection (AOI) equipment to spot defects on wafers as they are being processed. “Our AI driven Auto-Defect Classification system uses deep learning to sort and categorize millions of images every day,” says Mehrotra. “We also use thermal imaging in the fabs. We take images during normal workload conditions and overlay them on images from the fab’s digital twin. These maps give AI systems a baseline against which to compare the infrared images.”

When the systems detect a deviation, they automatically compensate or disposition to ensure the highest levels of quality and yield in its production.

Edge computing also enables predictive maintenance for the aerospace industry by filtering and aggregating data collected in flight before transmitting to the servers on the ground. Sending selective data from the aircraft makes it feasible to do real-time analytics for predictive maintenance.

“Intelligence at the edge can help determine what data should and should not be transmitted,” says Mehrotra. “This minimizes network bottlenecks and establishes data analytics at the most cost efficient and latency appropriate location.”

Advancing Semiconductor Manufacturing Technology

Micron has long been a tenacious innovator in memory process, often surprising bigger competitors with breakthrough technology leaps that enabled the company to hold its own in a very competitive industry, according to Mehrotra. “This year, I am extraordinarily proud to say, we took a leadership position in both DRAM and NAND process technologies simultaneously for the first time in our history. Our 1-alpha DRAM and 176-layer NAND are the most advanced memory and storage technologies in production today,” he notes.

Micron’s 1-alpha DRAM is the smallest linewidth geometry memory available in the world today. And producing this technology requires highly specialized tools, utilizing precision materials, and operating in enormous ultra-clean fabs. Interestingly, this technology is produced in Taiwan, in Micron’s manufacturing operations in Taoyuan and Taichung.

“Our 176-layer NAND is a breakthrough in storage technology—the most technologically advanced NAND in the market, and Micron’s fourth generation of stacked design,” says Mehrotra. “To achieve the capacities that can store terabytes of your photos, videos and music, Micron’s flash technology stacks data cells in hundreds of layers, providing space for billions of bits of information on a single chip. And I am proud of these technologies not just from a competitive perspective, but because of what they enable in our end solutions for customers around the globe.”

Keep Pushing Technology

Mehrotra sees a huge opportunity for memory and storage to play a greater role, as Micron’s engineering teams work closely to influence architectures across platforms to ensure the most efficient and effective access to valuable data.

“In the end, the innovation we create is driven by the ideas and expectations of you—the end-customer,” he says. “So, my call to each of you is keep dreaming. Keep inventing. Keep pushing technology to make life better for all. It is your ideas that drive our industry to break through barriers. With your help, we can take these powerful new technologies and create a safer, more comfortable, and more just world for all.”


Stephen Las Marias is the editor of EETimes Asia. He can be reached at stephen.lasmarias@aspencore.com.


Subscribe to Newsletter

Leave a comment