Home Articles Flash memory supply

Flash memory supply

How to address the challenges of sourcing Flash memory

Sourcing difficulties in the electronics industry are not new. Every year, natural disasters such as typhoons, hurricanes, and earthquakes affect the semiconductor manufacturing and assembly industry around the world. While the human costs are being assessed and response teams are tasked with providing support and assistance, the buying and selling teams need to work together to determine the amount of business impact. Due to the great specialization of each manufacturing plant, often only certain semiconductor or packaging devices are affected. The strength of the supply chain is such that those who provide components through distributors sometimes do not feel these effects. However, since the arrival of the coronavirus in early 2020 and the introduction of initiatives to combat its spread, there have been a series of chain reactions that have affected the supply of semiconductors significantly.

Why has this lack of supply occurred?

By introducing restrictions to limit people-to-people contact, many manufacturing companies were forced to close temporarily. At the same time, demand for high-end products such as private and commercial products declined as consumers began to reckon with economic uncertainty. Industries affected by these changes canceled semiconductor orders. At the same time, as everyone was learning and working from home, demand for products like tablets, monitors, and laptops skyrocketed. Semiconductor suppliers pushed their production capacity to the limit with these new orders to respond to such demand. As lockdown restrictions were relaxed, a situation arose where canceled orders were being reshipped to factories that were already operating at full capacity. It typically takes several months to produce complex semiconductor devices, so the wafer production process doesn't start unless there are enough orders planned for a given product. Currently, the sector is making an effort to prioritize manufacturing in order to satisfy as many customers as possible, but even so, it is not possible to supply everyone.

This has led purchasing teams and design engineers to collaborate to evaluate appropriate mitigation strategies. Highly specialized semiconductor devices, such as microcontrollers and advanced analog solutions, are difficult to replace. However, Flash memories, with their standardized pinouts and interfaces, seem to be a less complicated trade-off, at least in theory. However, as always, the difficulties lie in the finer details, and if a second-party device is not approved during the development process, it is unlikely that a simple device substitution will suffice. Even if an alternate Flash memory works on the first try, there may be hidden issues that lead to premature failure in the field due to greater than expected wear.

SLC NAND

SLC NAND Flash, available with both serial and parallel interfaces, provides a relatively high level of cross-vendor compatibility. On a physical level, the placement of pins and packs should be the same, although the solderability and suitability of the reels for placement equipment would still need to be reviewed. At the hardware interface level, Serial NAND uses SPI. Due to the different ways that the interface can be implemented in microcontrollers (MCUs) and systems-on-chips (SoCs), a basic test setup should be built to ensure access to flash memory. The same is true for parallel NAND Flash memory, where the timing of the signals may require adjustment using internal registers of the MCU or SoC.

And this brings us to the next challenge, which is the software. If the application code has followed good practice with low-level controllers when handling the interface, and with high-level controllers when handling the specific details of external memory and the file system, any changes to the code it should be reasonably simple. Tweaks may require additional logging support to be added or code to be changed to support a completely different logging implementation. Devices like Kioxia's NAND serial interface feature hardware error correcting code (ECC) support that can be turned on or off. The default setting of the ECC enabled, or the specific method of turning it off, may not match that of the device to be replaced. Devices like Kioxia's BENAND™ have integrated ECC capable of 8-bit error correction and 9-bit error detection. However, the parallel SLC NAND of Kioxia's latest 24nm technology node relies on a host processor to generate 8-bit ECC for each 512-byte memory block.

When 5 or 6 wrong bits are detected, the firmware takes care of rewriting the entire block. Needless to say, this process increases memory wear even when no application-level write operations have been performed on the software. The engineering team will need to carefully compare the data sheets of the initially chosen Flash memory with the replacement solution to understand how these additional memory writes will affect the lifespan of the flash storage.

Managed Flash: e-MMC and UFS

With the highest level of standardization in the world of managed flash devices, switching vendors for e-MMC and UFS storage is a little easier. JEDEC's JESD84-B51A standard, version 5.1a, defines the characteristics and electrical interface of the e-MMC. Coming to market in 2019, it is not expected to undergo any further changes in the future, reducing the likelihood of physical interoperability issues between SoCs and storage.

e-MMC devices offer separate write and delete cycle registers for their "boosted" (pSLC) and "normal" (MLC/TLC) blocks. However, these “health status” records only increase in 10% steps, which is not very granular.

Furthermore, this information only provides guidance as to memory health. Reading “100% used” in this log does not mean that the Flash device is not working, any more than a cheap brand tire does not indicate an accident is about to occur. It simply indicates that disaster is more likely.

The UFS is more modern and has the JEDEC JESD220E standard, also known as version 3.1, available since the beginning of 2020. Thanks to its significantly higher performance, it has quickly established itself as the storage option of choice in mobile and is gaining market share in the automotive sector (figure 1). At a time when developers want to learn more about the health of their storage to improve the user experience, it is expected that the standard will continue to be updated to respond to these needs.

automotive applications
Figure 1: UFS is becoming the preferred alternative to e-MMC in mobile and automotive applications due to its superior performance.

Although an initial physical swap of Flash storage media can produce good results, it is necessary to understand how the internal controller handles data. While the application can only perform a certain number of data writes, the Flash driver can cause additional data removal and page writes by attempting to reorganize unused dead space (Figure 2). The disparity between application writes by the host SoC and the writes made to the raw flash cells is known as the Write Amplication Factor (WAF). A perfect WAF would be 1, but a good target value for a typical application is 4 (figure 3).

flash driver
Figure 2: The Flash driver deletes at the block level and writes at the page level to clear up unused dead space, causing a mismatch between Flash cells and application writes.
host processor
Figure 3: The difference between the data written to Flash memory by the host processor and the number of writes to the NAND memory cells performed by the Flash controller is defined as WAF.

The exact way that host writes are translated into Flash cell writes depends on the Flash storage vendor. So while the workload may have been assessed during the purchase of the original Flash storage solution, the testing process will need to be repeated with the chosen new device vendor.

Seek support and get it fast

Due to the high level of compatibility at the pin and packet level, and a standardization or similarity in the Flash storage interconnect, it can be quickly assumed that providing an alternative device will be straightforward. However, the reality is very different. Standardization simplifies the process considerably, but you need to understand subtle hardware differences, which are not always visible on device data sheets. In the event of supply issues in the future, it is highly recommended to contact Flash providers to discuss your needs as soon as possible. Engineering teams will be able to analyze workload traces and provide guidance on necessary changes when changing storage device vendors. In addition, the commercial teams can also advise you on delivery times. Since Flash storage's lead time is months, suppliers will appreciate having advance information on their demands to ensure you receive the products you want when you need them.