What’s Next For NOR Flash?

The flash memory market is the tale two of cities.

Today, NAND and NOR are the two main flash memory types. Over the years, the NAND flash market has exploded. Targeted for data storage, NAND flash has moved into flash cards, solid-state storage drives (SSDs) and other products. The excitement for NAND continues to mount, as the technology is moving from planar to a 3D structure. In fact, 3D NAND is still in the early stages and the market is also exploding.

Amid the boom for NAND, NOR got lost in the shuffle. Typically, NOR is used for code storage. (One Web site, Radio-Electronics.com, provides a good tutorial regarding NAND and NOR.)

“The NOR market is a shadow of its former self,” said Jim Handy, an analyst with Objective Analysis. “It was up around $3.5 billion a few years ago, and is now around $1 billion. It lost out when the cell-phone market went to smartphones, which don’t use NOR. The old standby applications still use it: set-top boxes, PC BIOS, candy bar phones and others. But the unit demand hasn’t grown very fast for those applications and the prices have declined an average of 30%/year, so the revenues have declined significantly.”

Still, NOR flash is used for a multitude of applications. Today, the NOR market can be divided into two segments—standalone devices and embedded applications. Cypress, Macronix, Micron, Winbond and others sell standalone NOR devices.

Of the two segments, embedded NOR is arguably the most dynamic market. The embedded chip market itself is exploding on several fronts, such as automotive, industrial, medical, wireless and others.

For embedded applications, OEMs use microcontrollers (MCUs) and other devices. “The highest-volume MCUs tend to use old processes like 90nm and 130nm,” Handy said. “It will be a long time before many start to use the sub-20nm processes that require finFETs.”

MCUs integrate several components on the same chip, such as a CPU, SRAM, embedded memory and peripherals. The CPU executes the instructions. SRAM is integrated on the chip to store data or frequently used instructions. Then, embedded memory, such as EEPROM and NOR flash, are used for code storage and other functions. “The difference is whether it has one transistor per bit cell (NOR) or two (EEPROM). Most MCU makers use EEPROM and others use NOR,” Handy said.

Embedded NOR is robust. “It does one job really well,” said Dave Eggleston, vice president of embedded memory at GlobalFoundries. “It stores data in a harsh environment.”

But as systems become more complex, OEMs require embedded memory with more capabilities. With that in mind, some are asking a simple question—Is embedded NOR flash running out of steam? Answer: Embedded NOR remains viable, but there are challenges to scale the technology.

The mainstream market for embedded NOR flash is at 40nm and above, although the industry is beginning to migrate towards smaller geometries, said Walter Ng, vice president of business management at UMC. “The focus on the development side is on 28nm,” Ng said.

Embedded NOR has some limitations, however. Write speeds are slow. It requires more masks at each node, thereby impacting cost and complexity. “The biggest challenge of embedded non-volatile memories is cost reduction when moving to smaller geometries,” said Vipin Tiwari, director of worldwide marketing and business development at Silicon Storage Technology (SST), the embedded memory subsidiary of Microchip. “The traditional embedded flash cell continues to scale down well all the way to 28nm; however, high-voltage transistors do not scale in proportion. Therefore, the overall flash macro size doesn’t scale with Moore’s law. However, it is still desirable to port an embedded flash based product to smaller geometries due to ever increasing complexity and size of instruction code, which in turns requires significant increase in the size of embedded flash.”

The mainstream embedded flash technology from SST is 40nm, but the company is working on 28nm. Based on a proprietary split-gate flash memory cell, the company’s embedded flash is called SuperFlash. “Automotive applications, for the most part, have been driving embedded SuperFlash memory scaling at advanced technology nodes,” Tiwari said. “We expect the same trend to continue at the 28nm technology node as well. Regarding new types of applications, we are also expecting some of the high-end IoT devices that need on-chip RF integration to take advantage of embedded flash availability at 28nm.”

Tiwari and others believe that 28nm will be a long-lasting node for embedded flash and chips in general. But can NOR scale beyond 28nm? “We believe we can further scale the SuperFlash memory technology and integrate it in a FD-SOI technology platform as well,” he said. “We haven’t yet put a significant R&D effort on integrating SuperFlash memory technology with finFET structures on the 1xnm node, as there is no customer demand yet. We think it is going to be challenging to integrate a traditional floating-gate based technology with a finFET platform at a 1xnm node, and we might need an alternative solution.”

What’s next?
To be sure, the automotive industry would need an embedded flash memory that can scale. “Therefore, we expect that some of the automotive applications would move to a SIP solution with SuperFlash memory fabricated at the 2xnm or 4xnm nodes,” he said. So, instead of scaling embedded NOR to 16nm/14nm and beyond, Microchip’s SST sees a scenario where NOR is integrated into a system-in-package (SIP) scheme. The MCU could be part of the SIP as well.

Still others plan to scale embedded memory. At IEDM, for example, Renesas presented a paper about embedded flash technology for finFETs at 16nm/14nm and beyond. This technology may not appear for several years, however.

Over the years, though, Renesas has developed MCUs based on a proprietary flash technology called SG-MONOS. It involves a metal-oxide-nitride-oxide silicon (MONOS) technology with a charge-trap storage scheme.

For 16nm/14nm and beyond, the industry has talked about a traditional one transistor cell structure. In contrast, Renesas has proposed a 1.5 transistor cell structure. “Superior sub-threshold characteristics and small threshold-voltage variability owing to the fin-structure (have been demonstrated), which are required for future low power operations,” said Shibun Tsuda, a researcher from Renesas, in the IEDM paper. “Incremental step pulse programming for source side injection is found to be quite effective to suppress the fin top corner effect, and reliability improvement by homogenized electric field and electron injection is confirmed. Excellent retention in 150 ºC after 250K P/E cycles shows this device has sufficient reliability for advanced automotive applications.”

While embedded NOR continues to make gains, the industry is developing several next-generation memory types. Many are targeted to replace traditional embedded NOR. “Emerging memories such as MRAM are promising, and they can enable new applications; however, we have a long way to go before emerging memories can significantly reduce overall cost, compared to traditional embedded flash,” Microchip’s Tiwari said. “Number of masks is not an effective way to compare the cost of non-volatile memories.”

Embedded MRAM is gaining steam, however. For example, GlobalFoundries and Everspin are working on an embedded MRAM technology for GlobalFoundries’ 22nm FD-SOI platform. GlobalFoundries “plans to replace embedded NOR with MRAM for tight processes– anything with a finFET. Seems that they don’t expect finFET NOR to be developed,” Objective Analysis’ Handy said.

In addition, Samsung has fabricated an embedded 8-Mbit STT-MRAM within its 28nm process. “(The device proves the) feasibility of eMRAM commercialization for IoT applications,” said Y. J. Song, a principal engineer with Samsung.

Another technology, ReRAM, is also emerging for embedded applications. For example, in 2013, Panasonic rolled out an 8-bit MCU. The MCU was integrated with a 0.18-micron ReRAM technology. Recently, Panasonic formed an alliance with Taiwan’s UMC. Under the plan, the companies will develop a 40nm ReRAM process. UMC will manufacture the devices on a foundry basis for customers.

Panasonic is expected to ship product samples based on UMC’s 40nm process in 2018. The technology is aimed for both standalone and embedded applications. “PSCS was the first in the industry to mass produce ReRAM,” said Kazuhiro Koyama, president of Panasonic Semiconductor Solutions. “This cooperation with UMC will extend the range of optimal products that meet customer needs by developing a scaling process platform that will accelerate the market adoption of ReRAM.”

Related Stories
What’s Next For NAND?
Vendors pin hopes on 3D memory, raising specter of 2D NAND shortages; other memory types add to uncertainty.
Sorting Out Next-Gen Memory
A long list of new memory types is hitting the market, but which ones will be successful isn’t clear yet.
New Embedded Memories Ahead
Options grow as new wave of MCUs demand more capable NVM.
Tech Talk: Embedded Memories
Which type of memory is better and why.
The Future Of Memory (Part 1)
Experts at the table, part 1: DDR5 spec being defined; new SRAM under development.