Weaknesses of NOR Flash

NOR Flash is a popular choice for code storage in embedded systems. NOR Flash technology is inherently robust, possessing “end-to-end” signal integrity and supporting over 100,000 hours of data retention. NOR flash has cost competitiveness per bit in the capacity range up to 512 Mbits, but it lags far behind in terms of scaling (process miniaturization) advocated by Moore's Law.

However, NOR Flash's slow scaling poses a problem for many developers who are beginning to implement AI in their embedded systems. Applications such as machine learning generate complex codes, requiring code storage of 1 Gbit or more, and the unit price per bit of NOR flash is higher than that of NAND flash in this capacity range. This is because NAND flash followed Moore's Law from 46 nm, 32 nm, 2x nm, 1x nm. Since there is a very close correlation between chip area and cost, NAND flash is cheaper than NOR flash at smaller process nodes in the high density range of 1 Gbit and above.

Today's smart and connected devices require field updates and over-the-air (OTA) updates for security patches and feature upgrades. A typical OTA update overwrites existing code stored in non-volatile memory with new code. This means that the system must be powered down while the update is running. To minimize downtime, developers will want to overwrite updates as soon as possible. Therefore, the key performance for OTA updates is the program/erase time, which is better for serial NAND flash than serial NOR flash.

NAND Flash Reputation Problems

In AI applications for embedded systems, NAND flash has advantages in price and performance. However, in order to get developers to choose serial NAND, we need a mindset change. This is due to the preconceived notion of all NAND flash based solely on ultra-high-capacity NAND use cases.

In ultra-large capacity NAND flash for SSDs (solid state disks) used in laptop computers and tablets, data integrity and data retention are given lower priority to achieve large capacities and low unit prices per bit. In fact, corrupting or losing a few bits in music or video files is acceptable for ultra-high-capacity NAND flash manufactured at leading-edge process nodes. However, the performance of serial NAND flash optimized for code storage in embedded systems is significantly different from state-of-the-art ultra-high-capacity NAND flash.

Winbond High Performance and Reliability Serial NAND Flash

To accelerate the transition from serial NOR flash to serial NAND flash for code storage in embedded systems, Winbond has made NAND flash improvements as follows:

• 1 bit error per sector
• Fast read performance up to 83MB/s
• Hardware/software compatibility with serial NOR flash

The high reliability of Winbond's serial NAND flash QspiNAND (Quad SPI NAND) comes from its 46nm process manufacturing and SLC (Single Level Cell) memory cell configuration. This generation of manufacturing process has been in the market for many years and has proven its reliability and quality. Winbond's 46nm process serial NAND flash QspiNAND guarantees over 100,000 hours of data retention. In addition, Winbond's serial NAND flash QspiNAND implements 1-bit ECC (Error Correction Code) to maintain data integrity during both write and read operations.

Winbond 2nd Generation Serial NAND Flash with Enhanced Performance

Winbond's serial NAND flash QspiNAND offers superior performance and cost advantages over serial NOR flash. Even in AI implemented in embedded systems, inference engines implementing machine learning algorithms locally must frequently perform highly complex computing operations in milliseconds. This requires fast data read performance.

Figure 1: Winbond's second-generation serial NAND flash QspiNAND W25N-JW offers read throughput up to 83MB/s (Image credit: Winbond Electronics Co., Ltd.)

Winbond's first-generation serial NAND flash QspiNAND had a read throughput of up to 52MB/s. Winbond's second-generation serial NAND flash QspiNAND W25N-JW series, announced last year, has a maximum read throughput of 83MB/s (see Figure 1). Furthermore, the W72N-JW series, which stacks two W25N-JW chips, can double up to 166MB/s. The W72N-JW series consists of 8 I/O with dual Quad SPI interface. The host controller can operate the W72N-JW via one chip select pin (see Figure 2).

Figure 2: Winbond dual QspiNAND W72N-JW provides read throughput up to 166MB/s (Image credit: Winbond Electronics Co., Ltd.)

This fast read throughput reduces latency in embedded systems. Winbond's Serial NAND Flash QspiNAND also supports speedy OTA updates and has the performance of minimizing downtime. Serial NOR flash has a write throughput of 0.36MB/s, while serial NAND flash QspiNAND has 8.5MB/s. The total time to program 1 Gbit of data for serial NOR flash is about 6 minutes, while the second generation serial NAND flash QspiNAND takes only 15 seconds. The erase time is 150ms for a 64Kbyte block for serial NOR flash, but 2ms for a 128Kbyte block for serial NAND flash QspiNAND.

Easy integration into embedded systems

The trend to integrate AI functionality into embedded systems is driving the move to serial NAND flash as an alternative to costly serial NOR flash at densities of 1 Gbit and above. Winbond's serial NAND flash QspiNAND is software compatible with the interface of serial NOR flash, requiring only 5 additional commands for NAND flash specific operations, ECC and look-up table (LUT) control . In addition, it comes in an industry-standard footprint and compatible pinout, making it a simple replacement for existing designs implementing serial NOR flash.

Additionally, Winbond's Serial NAND Flash QspiNAND is supported by SoC providers such as NXP Semiconductors, STMicroelectronics and Renesas Electronics, whose ecosystem will drive further adoption. For example, NXP Semiconductors has adopted Winbond's SpiStack (serial NOR flash and serial NAND flash stacked in one package) for its FRWY-LS1012A development board for its edge computing processor LS1012A. The QspiNAND flash contains the Linux® operating system code and the serial NOR flash contains the boot code.

The 2nd generation QspiNAND is currently available at 1Gbit, but can be expanded to densities such as 2Gbit and 4Gbit. This provides a roadmap for the needs of embedded developers who require more than 1Gbit of code storage, such as with cutting-edge AI technology.

Inquiry

If you have any questions regarding this article, please contact us below.