Menu
Sanjeev Aggarwal, CEO, Everspin Technologies & the Non-Volatile Memory Markets
In the embedded memory space, STT-MRAM is replacing NOR Flash, and it is also bringing new levels of performance to SoC and ASIC designers quotes Sanjeev Aggarwal | CEO | Everspin Technologies Inc
Niloy Banerjee March 16, 2023
5 minutes read
In the embedded memory space, STT-MRAM is replacing NOR Flash, and it is also bringing new levels of performance to SoC and ASIC designers quotes
Sanjeev Aggarwal | CEO | Everspin Technologies Inc. while in an interview with Niloy from BISinotech. In this interview the veteran underlines various aspects inking the future of memory technology including MRAM, STT-MRAM, the potential of neuromorphic computing. Also he highlights Everspin’s dominance and growth trajectories in the coming fiscal years. Edited Excerpts Below.
As MRAM moves into the embedded space, MRAM is also known to emerge as a persistent memory for numerous applications. How instrumental will Everspin be in driving the growth in this space?
Everspin has been producing STT-MRAM for more than 5 years. The first products we introduced to the market were persistent DRAM-like products, starting with our 256Mb DDR3, and following that with the 1Gb DDR4. Since then, we have optimized our STT-MRAM technology to support the persistent SRAM-like products for more extreme requirements of the industrial electronics market. Last year we announced volume production of the EMxxLX family of xSPI STT-MRAM that supports industrial temperature range and has longer data retention with unlimited write cycle endurance. The EMxxLX is compatible with
NOR Flashserial memories but also offers the performance of Static RAM with byte addressability. This product opens up the NOR Flash market for customers needing faster writes and lower power.
STT-MRAM finally comes to market replacing embedded NOR Flash. Everspin’s dominance in this growing market?
In the embedded memory space, STT-MRAM is replacing NOR Flash, and it is also bringing new levels of performance to SoC and
ASIC designers. NOR Flash has not scaled well below 40nm CMOS nodes and STT-MRAM has stepped in to fill the gap. Everspin has licensed its STT-MRAM technology to GlobalFoundries and partnered with them to offer embedded STT-MRAM at 22nm and below. Customers can now integrate this into their own designs getting the benefits of non-volatility but with faster write speeds and higher write cycle
endurance compared to NOR Flash.
What are the challenges confronting the fabrication and testing of STT-RAM?
Emerging technologies in early stages are typically challenged with manufacturing costs and availability of specific manufacturing tools, metrology tools, and maturity that impact its adoption rate in the industry. Everspin has been in volume production of STT-MRAM since 2017 at GlobalFoundries improving manufacturing efficiency
and yields. Manufacturing tools providers now offer MRAM-specific tools which, in turn, enables volume production scaling and cost reduction. Further cost reductions will be a function of demand and adoption in the broader industry.
May it be connected cars, integration of ADAS or sublime autonomous vehicles, automotive electronics systems are hungry for memory solutions. How does Everspin look into this domain and key offerings to cater for this market?
As connected devices get smarter and more remote, the need to have data protected in harsh environments as well protected in the event of power loss becomes paramount. EVs may have thousands of data collection points including instrument clusters, cabin preferences, safety systems, battery monitoring, motor control, assisted driving and more. The subsystems controlling this typically have a need for a reliable, persistent memory to capture and store the data from these points. In many case, the data volume of these systems is not huge and can be handled by multiple MRAMs in the various subsystems. Examples of where MRAM gets used are as a personality” memory in infotainment and cabin preferences, BMS health monitoring, security alarm systems, event recorders. This is true in personal vehicles as well as commercial transportation modes such as trains, aircraft and the coming eVTOL market.
SRAM segment still has a viable dominance in the consumer electronics market such as smartphones and wearable devices. Do you see a technological shift in the coming time with MRAM replacing the existing SRAM segment?
SRAM is used broadly in electronic systems, both as discrete and embedded memory. In the industrial segment, SRAM is often backed up with a battery and this is where MRAM has excelled due to a combination of non-volatility and random-access performance. In the consumer electronics segment, particularly in wearable applications, SRAM suffers from relatively high leakage currents in the advanced CMOS process nodes which affects battery life. This is where MRAM can play a huge role with an inherently lower leakage memory cell,
but also a “zero standby power” capability because the MRAM can be totally shut down without losing its data. There is a big effort underway by the foundries to refine eMRAM for the performance of caching for processors.
This industry has been spectating for joint ventures and collaborations with other manufacturers to create a sophisticated and broad value chain also to have a competitive advantage over other competitors. What has been Everspin’s strategy to be competitive in the market?
Our ability to compete comes from the unique combination of practical MRAM research with a learned know-how of manufacturability for commercial success in MRAM products. Establishing partnerships and collaborations is necessary to be able to scale up the production in advanced process nodes and to continue to make the process and materials advance that drive down cost. We have a rich history of partnering, having collaborated with GlobalFoundries since 2014 to develop and produce STT-MRAM on 300mm and advanced CMOS nodes for several generations. The embedded MRAM offering from foundries, including GlobalFoundries, are driving wider adoption of MRAM as it becomes a preferred technology over Flash and SRAM. Partners have licensed our technology and augmented the inherent capability of MRAM to make radiation-hardened memories. Recently
we have announced an effort with QuickLogic to develop MRAM technology for use in high-reliability
FPGA technologyin a DoD-sponsored program. Everspin is at the forefront of enabling the broader ecosystem for MRAM manufacturing and bringing the technology to additional value-added markets.
Everspin Technologies Inc. is known to be planning to expand its domestic MRAM manufacturing in the state of Indiana. Kindly tell our readers about these strategic expansion plans.
Everspin is working to build a trusted manufacturing line in the state of Indiana to add domestic capacity for Everspin’s Commercial MRAM and increased capability to act as a foundry for the manufacture of Toggle and STT-MRAM. Plans include working with the local research community to enhance domestic research for MRAM technology development, creating a Technology Development Center at the proposed Indiana-based location.
From Press Coverage:
SIBR 1/19/23: Westgate One will be a combined effort of NHanced Semiconductors, Everspin Technologies, Trusted Semiconductor Solutions and Reliable MicroSystems. The four semiconductor companies plan to invest, with assistance from the U.S. government, up to $300 million in the campus.
In recent years, neuromorphic computing has emerged as a promising technology in the post Moore’s law era. Your comments on this promising technology.
Replicating the processing power of the human brain has led to research in highly distributed memory systems that both compute and transmit data in attempts to mimic the synapses of the brain in massively parallel neural networks. This is generally in a research phase and MRAM has real potential to be a core technology in this
field. MRAM can be developed to provide multiple states as opposed to the traditional binary 1 and 0 of conventional memory. This can be architected in arrays that provide new abilities in AI learning and inference to modify weights in tables very rapidly and maintain the states until they change. In collaboration with Prof. Joseph Friedman at the University of Texas, Dallas Everspin has done an experimental demonstration of a neuromorphic network with STT magnetic tunnel junction (MTJ)synapses, which performs image recognition via vector-matrix
multiplication1. Practical implementations are being explored in which some processing logic is combined with the MRAM array in a common silicon chip, basically in-memory computing where table weighting or signal processing is distributed, not having to go back to the CPU for each step in a calculation. It will be an exciting new frontier for the industry and MRAM could very well be a key enabler.