Legacy memories are no longer lowly devices that hit their end of life (EOL) because a major vendor is focused on the latest and greatest...
Legacy memories are no longer lowly devices that hit their end of life (EOL) because a major vendor is focused on the latest and greatest. With a growing list of smart devices, AI-infused edge computing and the exponentially growing Internet of Things (IoT), “enduring” might be a better label for these entrenched memories.
Even with a DDR5 specification now available and emerging memories such as Intel’s Optane 3D Xpoint memory clearing the adoption hump, ever-changing economics still mean there’s a need for older, low-density memories, including older DRAM.
“In bygone days, DRAM was not used in smaller systems because they required a DRAM controller,” said Jim Handy, principal analyst with Objective Analysis. SRAM went into systems with small memory requirements; DRAM into the larger ones, accompanied by a controller used for larger memories.
“Moore’s Law has shrunk the cost of that controller down to nearly nothing,” Handy added.
While this has reduced the opportunities for SRAM, even though it’s appealing for high-speed systems, there remain plenty of applications requiring anything faster than SDRAM, DDR and DDR2 DRAM, the analyst said. LPDDR has solved some of the power problems that originally kept DRAM out of many portable applications.
“It used to be that SRAM was for low densities, low power and fast applications, but DRAM’s improvements have nibbled away at those markets,” Handy said. “This means that there is now a vibrant market for low-density DRAM that previously didn’t exist.”
Major memory vendors will continue to pursue leading-edge densities because they are fine-tuned to build hundreds of millions of identical parts to fill billion-dollar orders. “It’s less efficient for them to go after smaller opportunities,” Handy explained. “Yet, there are lots of these smaller opportunities out there, and they can be more profitable than the bigger part of the business.”
Hence, he predicts legacy memories companies such as Alliance Memory and others can still turn a tidy profit churning out commodity memory devices.
It was the EOL-ing of still-viable memories that prompted CEO David Bagby to launch Alliance Memory, having worked at Samsung, which recently announced it was exiting the DDR2 DRAM business.
Bagby said it’s not entirely clear when a memory falls into the “legacy” category. So far, it’s yet to include DDR3, but really comes down to each supplier’s strategy. In the last few years, for exampke, Micron has started to follow the Samsung model of EOL-ing older memory products.
Bagby said mid-size customers track the direction of the “Big Three” memory makers: Micron, Samsung and SK Hynix. “They don’t want to be stuck without being able to get a supply.” That’s Alliance Memory’s niche, providing customers with a reliable second source and fallback as they continue to using higher-density memories. When these vendors tag a memory as “EOL” as they focus on fewer, yet bigger customers with fewer product numbers, Alliance can continue to deliver.
Micron’s decision to stop production of some memory types creates opportunities for smaller players, Bagby noted. Alliance, is adding more products across memory types, including DRAM, SRAM and NOR flash.
Meanwhile, the company is also meeting demand for 4G and 8G DDR4, for example. “We’re in the mainstream. Customers are wanting DDR4 because of the processors.”
Alliance Memory is now an authorized reseller for Micron’s M29F 5V parallel NOR flash devices, which Micron has discontinued. It will continue to produce devices for Alliance, but will only fill new orders placed through mid-2021. Beyond that date, there’s an opportunity for the reseller to shift customers to an Alliance part.
Legacy memory vendors offer predictable pricing and long-term customer support, important since new designs still use older memories. For example, early DDR, SRAM and NOR flash are being used in medical, industrials and smart home applications.
“We’re getting way more new designs on DDR3,” Bagby said.
Demand for DDR3 memory is growing for non-smartphone and -tablet designs such as medical devices and industrial platforms. “We’re growing in SDRAM, we’re growing in DDR1 and we’re going to have a huge growth on DDR2 because of the Samsung EOL,” said Bagby.
Demand for legacy memories doesn’t always stem from memory replaced in older equipment; there are processor vendors targeting medical and industrial systems that are still designed around DDR3 technology. That equipment doesn’t require DDR4, and vendors like Alliance Memory don’t want to compete with larger vendors. According to Bagby, “That’s why they’re developing different processors with different memory requirements.”
Legacy memory technology doesn’t preclude innovation, a reason companies like Cypress Semiconductor are wary of the “legacy” label, preferring instead to call earlier memory specifications “enduring.” Even as high performance computing, hyper-scalers, AI and machine learning are driving adoption of the latest DDR memory, NAND flash and High Bandwidth Memory, it’s data that’s driving innovation, said Sandeep Krishnegowda, marketing and applications director of the Flash Business Unit at Cypress, an Infineon unit. “When you get to the Internet of Things with its sensor and edge nodes, requirements haven’t changed all that much,” he said. “People still want something that’s low power. People still want something that’s reliable [and] works as it’s supposed to.”
That demand is reversing the decline of technologies such as NOR flash, a memory technology gaining a new lease on life in automotive applications, said Krishnegowda. IoT and other emerging sectors like wearables are also generating requirements for low power memories with tiny footprints that are also reliable and secure.
The network edge is also creating new memory applications, added Cypress product marketing engineer Doug Mitchell, including healthcare and distributed IoT sensor nodes for both consumer and industrial applications. Rather than gigabytes of memory, billions of IoT sensor nodes will instead require highly optimized, purpose-built memories for IoT applications, including NOR flash, traditional asynchronous SRAM or FRAM. “You optimize those technologies for the use case, for the application, and that optimization occurs by minimizing power or minimizing footprint or getting the right density size for that application,” Mitchell said.
Memory technologies also can be tweaked for emerging applications. “You can create whole new product portfolios,” said Mitchell. “That’s exactly the kind of business we’re attacking. We have purpose-built memories that go into these specific applications.”
Edge computing, 5G infrastructure and automotive all have requirements for low power memories like NOR flash, reinforcing Cypress’ “enduring” memory pitch.
Macronix sees similar opportunities for memories considered legacy, including 5G and automotive applications. Anthony Le, the company’s senior marketing director, said the death of a NOR flash has been predicted for a decade, but the company continues to develop architectures supporting NOR flash interfaces.
Demand is driven by reliability requirements for telecommunications, for example. “That’s one of our biggest businesses, and they are still using NOR flash because it’s still the most reliable non-volatile memory,” Le said. Even as the protocols change from 4G to 5G, a lot of the underlying hardware in terms of technology remain the same, said Le.
Then there is the reduction of risk inherent in the transition to a newer memory technology. “When you go to a new interface or a new technology, you’ve got to rewrite everything,” noted Le. Macronix also sees plenty of opportunities to innovate around small density memories, not just because they’re cost effective on a per-bit basis, but also because many use cases require a small footprint even as densities grow and interfaces run faster.
Automotive applications in particular require qualified memory devices that will last a long time in an embedded environment subject to extreme temperatures. Newer memories are simply too expensive for automotive, Le said, or the required densities aren’t available in the right packaging. “That’s why NOR is still a sweet spot for automotive.”
There can be as many as four to six radar platforms in a car, each requiring memory to boot up. That memory must withstand harsh environmental conditions. NOR is preferred for its longevity. “The technology is solid, and we can provide longevity in terms of supply to communication, automotive and industrial—even medical,” said Le.
Medical applications also don’t require reinventing the wheel, even with new 5G wireless protocols. Designers are “not adding new interfaces for memory. The only thing that is being bolted on is communication modules,” Le explained. With regulatory approvals taking years, he said, “Once you build a software stack for an MCU, the next version of that MCU has to be pretty damn close because you don’t want to requalify that software stack.”
It’s not just the track record of the device itself that makes these enduring memories appealing, added Krishnegowda of Cypress. “It’s the entire supply chain [delivering] the product. If you’re counting on having products in automotive applications for the next 10 years, you better have somebody who has a complete supply chain.”
While memory technologies expand into new realms like persistence, so too do legacy products remain relevant, a function of their simplicity, reliability and longevity.