On June 4, 1968, the DRAM patent filed by IBM’s Robert H. Dennard was granted. To commemorate this anniversary and highlight the evolution of the technology, we have asked some of the manufacturers in our portfolio about their track record in DRAM and how they view the future of the technology. One of them is Longsys, a technology-based memory manufacturer that has been founded in 1999 and offers a broad range of DRAM and flash products for industrial customers with its FORESEE brand.
When did you enter DRAM?
Longsys is always working on providing complete memory products combination, we had already built up a full series Flash product line, but realized that a completed DRAM product line was necessary to offer better service for our customers.
So Longsys entered DRAM in 2010 and the FORESEE LPDDR product rage went into mass production in 2014 and now consists of the Foresee LPDDR4x, LPDDR5, eMCP, uMCP and ePOP. We began development and production of FORESEE module products in 2019.
As of today, Longsys owns FORESEE, an industry-class storage brand, and Longsys, an enterprise-class storage brand.
What milestones are you particularly proud of and why?
After entering DRAM in 2010, we are proud that our FORESEE LPDDR products went into mass production and shipments in 2014 and LPDDR4x shipments in 2018. The eMCP 6432 shipments followed in 2020 and the ePOP3/ePOP4x in 2022. In 2023 we will ship LPDDR5.
Another important milestone is the industrial grade DDR3 promotion ever since 2020, within two years we had achieved more than 60 Din cases. Industrial applications won’t easily adopt new DRAM products because of its harsh requirement on reliability, we had successfully verified our industrial grade DDR3 products on kinds of industrial applications which indicated the quality control capability and technical innovation capability of Longsys.
FORESEE module products started late, but in 2021 we achieved trial production and bulk production of RDIMM, which represents a rapid improvement in Longsys’s R&D capability. In 2022, we were the first to release DDR5 16GB UDIMM/SODIMM products, which also represented our rapid catch-up in the consumer category.
Lexar is the global consumer brand of Longsys, committed to high-performance, high-quality memory products. At present, Lexar launched ARES DDR5 6400 CL32, which is the gaming memory module with the leading indicators in performance, power consumption, heat dissipation and some related specifications.
In the future, Lexar will faster roll out more OC gaming memory module above frequency 8400. However, Lexar will continue to cultivate gaming memory and memory products.
What’s hot in DRAM for you now and why?
The hot technology of DRAM is HBM-like (high bandwidth memory) architecture DRAM products, by separating peripheral circuit from DRAM array circuit coupled with TSV technology (through-silicon via) new DRAM products would provide high bandwidth for embedded and PC applications, which would in a large extent improve the performance of all the smart electronic systems. New technologies especially for AI-related new technology HBM-like products are really necessary, it could be a key factor for new technology innovation.
Presently, the mainstream products are focussing on DDR5 DIMM, OC DIMM and embedded LPDDR5. Longsys has made a perfect layout in DDR4 and is ready to enter the diachronous DDR5 products. We have hundreds of senior engineers in Zhongshan Industrial Park, and have made considerable investment in R&D, testing and quality. Therefore, we have full confidence and technical capabilities in memory products to facing different application scenarios and customer groups.
Currently, the company has established long-term relationships with Micron and Hynix through strategic partnerships. Relying on self-developed automated DRAM screening equipment, the company has built its own plant in Zhongshan for capacity planning and layout.
What’s next in DRAM?
Like all semiconductor technologies, DRAM memory strives to get higher capacity and higher speed on a smaller form factor. So it’s not only since NAND started going 3D that the industry is looking at that option for DRAM as well. So far, HBM (high bandwidth memory) is an example for a 3D integration of up to eight DRAM dies. However, this process is very expensive as the memory is often connected to the memory controller or CPU through a special substrate that is costly. Additionally, testing of this type of memory is very difficult and expensive which is a hurdle to a broad adoption.
There are two other promising approaches, but these are still in early development stages. One is the 4F2 architecture that is defined as having a memory cell at each and every possible location, that being each and every crossing of Word Line and Bit Line, with the cell being 2F x 2F (and F being the minimum feature size). Memory manufacturers are working on this success of the current 6F2 cell structure for years, but the vertical transistors and floating body issues are still a barrier. However, recent research shows that this is achievable with Atomic Layer Deposition technique.
The other promising approach is a capacitor-less DRAM cell, so called “1T0C” that uses a floating body capacitance to store information instead. Another capacitor-less approach is “2T0C“ that also has similar benefits. They promise a simple structure for a simple device with good noise margins. There have been some advancements recently, but it still needs to prove it can add value without too much additional costs.
3D technology would be gradually adopted in DRAM devices with the innovation of 3D capacitor process, by incorporating the HBM architecture into DRAM chip design the performance and the density and the yield rate of DRAM chips would be improved to a large extent. 3D architecture DRAM could further pull up the DRAM speed performance relative to 2D DRAM by using multi-level buffer organization in DRAM chip.
Do you think DRAM will ever be replaced by a new technology?
DRAM has been so successful because the technology offers several advantages: low latency, an almost lifetime endurance, comparably low power consumption, and most importantly, low costs. Any technology that wants to challenge DRAM needs to provide these advantages, most of all the cost efficiency of achieving 16 billion 1T1C cells at single-digit $ costs. As we see it, DRAM is very much alive, and while all in the industry agree that there are drawbacks in the underlying Van Newman architecture, all emerging memories are complementing DRAM but not replacing it. So we see DRAM technology as still alive and vibrant and we will keep on investing in evolving our technology and improving DRAM products for our customers.
In the short term, like 5 years, there won’t be any new technology that could replace DRAM, because most electronic systems were built under the DRAM technology, and the organization of electronic systems won’t be easily and efficiently changed. In the long run, DRAM would definitely be replaced because the process of DRAM would meet limitations soon, it won’t be possible for DRAM product to keep shrinking, and without advanced process support the speed of DRAM would soon meet a bottleneck. In the future, some NVRAM memory (could be FRAM based or MRAM etc. based) products could replace DRAM by integrating DRAM and Flash together into one ‘mega-memory’ device to lift up the efficiency of system command execution.
