The Evolution of Memory Technologies: A 40-Year Journey
For four decades, the landscape of semiconductor memory has transformed dramatically. From the early days in the 1980s, when our choices were limited to SRAM, DRAM, EPROM, and non-Flash EEPROM, the evolution of memory technologies has been profound. As I transitioned from an engineer to an editor, I witnessed the emergence of innovative persistent memories, notably from companies in Colorado Springs such as Ramtron’s ferroelectric memory (FRAM) and Simtek’s SONOS Flash memory. While Ramtron is no longer in the picture, FRAM persists, and Simtek’s legacy continues under Cypress, which was acquired by Infineon in 2020.
The Rise of Flash and DRAM
Flash memory burst onto the scene when Toshiba began production in 1987, gaining traction in the 1990s. Today, Flash EEPROM is essential for both on-board and on-chip code and data storage. Meanwhile, DRAM has evolved into more complex forms like SDRAM, which introduced intricacies in RAS/CAS timing and refresh cycles while complicating circuit board design with demands for matched-length high-speed traces.
In recent years, we’ve seen several alternative and persistent memory technologies emerge, including updated versions of FRAM, magnetic RAM (MRAM), resistive memory (ReRAM), and phase-change memory (PCM). These technologies are striving to carve out their niches as persistent storage class memory (SCM), a long-elusive goal in the memory sector.
Insights from the SNIA Webinar
In January, a pivotal webinar organized by the Storage Networking Industry Association (SNIA) featured Tom Coughlin from Coughlin Associates and Jim Handy from Objective Analysis, both notable experts in memory technology. Their insights provided a contemporary status report on alternative memory developments, especially for embedded systems, which are becoming increasingly important.
Coughlin initiated the dialogue by addressing the memory solutions in microcontrollers. These ubiquitous chips primarily utilize NOR Flash for storing code and some SRAM for volatile data. As technology progressed, the scalability of NOR Flash plateaued at 28nm due to its last planar FET process node, necessitating a shift towards alternative nonvolatile memory solutions for future microcontrollers.
Alternatives to NOR Flash
Coughlin outlined three possible solutions for microcontrollers adapting to smaller manufacturing nodes. The first option is to combine FinFET logic circuits with planar NOR Flash on a single chip. However, this approach is economically unviable for low-cost components like microcontrollers, as it would mandate traversing advanced semiconductor processes that significantly inflate costs.
The second alternative proposes using external Flash storage for code, complicating package design as additional pins would be necessary for communication with external memory. This setup is often more feasible for larger FPGAs than for microcontrollers due to pin limitations and elevated costs.
This brings us to the third option: implementing a new on-chip storage technology. Here, candidates like FRAM, MRAM, and ReRAM are making strides. Notably, NXP and TSMC are collaborating on MRAM-based microcontrollers for automotive applications, and Texas Instruments’ MSP430FR57xx family already integrates FRAM for non-volatile storage.
Transition to Alternative Nonvolatile Memory
Jim Handy then delved deeper into the foundational memory technologies shaping semiconductor evolution since the 1970s. NAND Flash memory faced significant hurdles scaling beyond 15nm, necessitating a shift to 3D NAND technology, which has led to astonishing achievements like SK hynix’s 321-layer chip. Yet, even these complex structures have their limitations as they too may halt scaling at 10nm.
Handy emphasized the growing interest in alternative memory technologies, which appear more promising for future device scaling. The initial push for implementation, however, has encountered setbacks. The Intel and Micron collaboration on PCM, branded as Optane and 3D XPoint, was ultimately discontinued due to the high production costs compared to traditional NAND Flash and DRAM.
Nevertheless, the potential of these innovative memory technologies aligns them with the continued miniaturization of semiconductor chips, allowing for a more efficient integration that confines costs.
Advantages and Anticipated Changes
All alternative nonvolatile memory technologies exhibit appealing characteristics, particularly radiation tolerance, which is crucial for military and space applications. These new memory types also write data significantly faster than traditional Flash memories and allow for fine-grained byte-writing, which streamlines performance since entire pages no longer need to be erased and rewritten.
Despite these advantages, Handy pointed out that the requirement for innovative materials could pose a significant barrier to mass adoption.
A Look Ahead: Timelines for Adoption
Towards the end of his presentation, Handy presented a rough timeline suggesting that while MRAM is already utilized in small applications such as hearing aids, it could take another decade before these new memory technologies replace Flash memory in broader embedded applications. The slow pace of development in microcontrollers further complicates the transition, as cost considerations currently favor established NAND and SDRAM technologies.
Handy anticipates that once the shift begins, it will happen swiftly, impacting both embedded and standalone memory chips. The advent of memory chiplets is also on the horizon, though he cautions that their adoption will proceed at an even slower pace than that of alternative memories.
The Deep Dive into New Memory Technologies
For those seeking to explore these advancements further, Coughlin and Handy have collaborated on an extensive report titled “A Deep Look at New Memories,” available for purchase. Additionally, the SNIA Webinar can be accessed for an informative overview of the latest developments in memory technology.
As the digital landscape continues to evolve, the developments in alternative memory technologies will undoubtedly shape the future of computing, making the following decades an exciting time for innovations in this crucial field.