flash memory

Technical characteristics

NOR type and NAND type flash memory are very different. For example, NOR type flash memory is more like memory, with separate address lines and data lines, but the price is more expensive and the capacity is smaller; while NAND type is more like a hard disk, address line It is a shared I/O line with the data line. All the information similar to the hard disk is transmitted through one hard disk line, and compared with the NOR type flash memory, the NAND type has a lower cost and a much larger capacity. Therefore, NOR flash memory is more suitable for frequent random read and write occasions. It is usually used to store program code and run directly in flash memory. Mobile phones are large users of NOR flash memory, so the "memory" capacity of mobile phones is usually not large; NAND flash memory It is mainly used to store data. Our commonly used flash memory products, such as flash disks and digital memory cards, use NAND flash memory. Here we also need to correct a concept, that is, the speed of flash memory is actually very limited, its operating speed and frequency are much lower than that of memory, and the operation method of NAND flash memory similar to hard disk is also much slower than the direct access method of memory . Therefore, don't think that the performance bottleneck of the flash disk is in the interface, and even assume that the flash disk will get a huge performance improvement after using the USB2.0 interface.

As mentioned earlier, the operation method of NAND flash memory is inefficient. This is related to its architectural design and interface design. It really operates like a hard disk (in fact, NAND flash memory did consider the compatibility with the hard disk at the beginning of the design), it The performance characteristics are also very similar to hard disks: the operation speed of small data blocks is very slow, while the speed of large data blocks is very fast, this difference is much larger than other storage media. This performance feature is very worthy of our attention.

Flash memory access is relatively fast, no noise, and small heat dissipation. The demand for user space capacity is small. If you plan to purchase, you can not consider too much, and buy flash memory for the same storage space. If you need a large capacity (such as 500G), buy a hard disk, which is cheaper and can meet the needs of user applications.

Classification

By type

U disk, CF card, SM card, SD/MMC card, memory stick, XD card, MS card, TF card, PCIe flash memory card

By brand

SIS, Kingston, Sony, LSI, SanDisk, Kingmax, Eagle, Transcend, Patriot, Newman, ADATA, Lenovo, Taipower, MSI, SSK, Samsung, Hynix

[NAND flash memory] The basic storage unit of memory and NOR flash memory is bit, and users can randomly access any bit of information. The basic storage unit of NAND flash memory is Page (you can see that the page of NAND flash memory is similar to the sector of the hard disk, and one sector of the hard disk is also 512 bytes). The effective capacity of each page is a multiple of 512 bytes. The so-called effective capacity refers to the part used for data storage. In fact, 16 bytes of verification information are added, so we can use the technical information of the flash memory manufacturer.

Sandisk

You can see the representation of "(512+16) Byte" in the data. The vast majority of NAND flash memory with a capacity below 2Gb has a page capacity of (512+16) bytes, and NAND flash memory with a capacity above 2Gb expands the page capacity to (2048+64) bytes.

NAND flash memory performs erase operations in units of sectors. The writing operation of the flash memory must be performed in the blank area. If the target area already has data, it must be erased before writing, so the erase operation is the basic operation of the flash memory. Generally, each block contains 32 512-byte pages with a capacity of 16KB. When a large-capacity flash memory uses 2KB pages, each block contains 64 pages with a capacity of 128KB.

The I/O interface of each NAND flash memory is generally 8, each data line transmits (512+16) bit of information at a time, and 8 are (512+16)×8bit, which is the 512 bytes mentioned above. But the larger capacity NAND flash memory is also more and more adopting the design of 16 I/O lines. For example, the Samsung number K9K1G16U0A is a 64M×16bit NAND flash memory with a capacity of 1Gb and the basic data unit is (256+8 )×16bit, or 512 bytes.

When addressing, the NAND flash memory transmits address information packets through 8 I/O interface data lines, and each packet transmits 8-bit address information. Due to the relatively large capacity of flash memory chips, a group of 8-bit addresses can only address 256 pages, which is obviously not enough, so usually an address transfer needs to be divided into several groups, occupying several clock cycles. The address information of NAND includes the column address (starting operation address in the page), block address and corresponding page address. When transmitting, they are grouped separately. It needs at least three times and takes three cycles. With the increase of capacity, there will be more address information, which requires more clock cycles to transmit. Therefore, an important feature of NAND flash memory is that the larger the capacity, the longer the addressing time. Moreover, because the transfer address cycle is longer than other storage media, NAND flash memory is less suitable for a large number of small-capacity read and write requests than other storage media.

[1] The flash memory products with larger storage capacity and faster speed than our usual U disks belong to PCIe flash memory cards. It uses low-power, high-performance flash memory chips to improve application performance. Because they are plugged directly into the server, the data location is close to the server's processor, which saves time compared to other disk-based storage network paths to obtain information. Enterprises are turning to this technology to solve storage intensive

Workloads, such as transaction processing applications. In terms of PCIe flash memory cards, LSI’s new Nytro product expands its flash-based application acceleration technology to enterprises of all sizes. LSI launched three products to a PCIe flash adapter card market that is becoming increasingly crowded. Part of LSI Nytro’s product strategy, LSI’s WarpDrive card uses flash storage, LSI’s SAS integrated controller, and SandForce’s technology from the company’s acquired flash controller manufacturer SandForce. Its second-generation PCIe-based application accelerator card capacity ranges from 200GB to 3.2TB. The Nytro XD application accelerates the combination of software and hardware for storage solutions. It integrates the WarpDrive card and Nytro XD intelligent cache software to increase I/O speed in storage area network (SAN) and direct attached storage (DAS) implementations. Finally, there is the Nytro MegaRAID application accelerator card, which combines the MegaRAID controller with onboard flash memory and cache software. LSI has positioned Nytro MegaRAID for the low end, a performance enhancement solution for serially connected SCSI (SAS) DAS environments .

Claude Lorenson, Microsoft's director of product management for SQL Server, is optimistic about the future of LSI's flash memory products in the Microsoft server environment. Because LSI's flash memory product Nytro MegaRAID can help Microsoft SQL achieve a 10-fold increase in transactions per second,

[1] "Flash storage technology, such as LSI's Nytro application acceleration product portfolio, can be used to accelerate business-critical applications, such as SQL Server 2012," Lorenson said in a company statement. "With Microsoft going into Windows Server 8. With the enhancements provided, the importance of these technologies will continue to grow."

Storage principle

To explain the storage principle of flash memory, let's start with EPROM and EEPROM.

EPROM means that the content can be erased by special means and then rewritten. The basic unit circuit (memory cell) is often a floating gate avalanche injection MOS circuit, referred to as FAMOS. It is similar to the MOS circuit, in which two high-concentration P-type regions are grown on the N-type substrate, and the source electrode S and the drain electrode D are respectively led out through ohmic contact. There is a polysilicon gate floating in the SiO2 insulating layer between the source and the drain, and there is no direct electrical connection with the surrounding. This kind of circuit indicates whether the floating gate is charged or not. If the floating gate is charged (for example, negative charge), just below it, a positive conductive channel is induced between the source and drain, so that The MOS tube is turned on, which means that 0 is stored. If the floating gate is not charged, a conductive channel is not formed, and the MOS tube is not conductive, that is, 1 is stored.

The working principle of the EEPROM basic storage unit circuit is shown in the figure below. Similar to EPROM, it generates a floating gate above the floating gate of the EPROM basic unit circuit. The former is called the first-stage floating gate, and the latter is called the second-stage floating gate. An electrode can be drawn to the second-stage floating gate to connect the second-stage floating gate to a certain voltage VG. If VG is a positive voltage, a tunnel effect is generated between the first floating gate and the drain, so that electrons are injected into the first floating gate, that is, programming and writing. If VG is made a negative voltage, the electrons of the first-stage floating gate are strongly lost, that is, erased. It can be rewritten after erasing.

The basic unit circuit of flash memory, similar to EEPROM, is also composed of a double-layer floating gate MOS tube. However, the gate dielectric of the first layer is very thin and serves as a tunnel oxide layer. The writing method is the same as that of EEPROM. A positive voltage is applied to the second-stage floating gate to make electrons enter the first-stage floating gate. The reading method is the same as EPROM. The method of erasing is to apply a positive voltage to the source and use the tunnel effect between the first-stage floating gate and the source to attract the negative charge injected into the floating gate to the source. Because the source plus positive voltage is used for erasing, the source of each unit is connected together. In this way, the flash memory cannot be erased by byte, but is erased in full or block. Later, with the improvement of semiconductor technology, flash memory also realized the design of single transistor (1T), mainly adding floating gate and select gate to the original transistor,

A floating shed for storing electrons is formed on the semiconductor where the current is unidirectionally conducted between the source and the drain. The floating gate is wrapped with a layer of silicon oxide film insulator. Above it is the selection/control gate that controls the conduction current between the source and drain. The data is 0 or 1 depending on whether there are electrons in the floating gate formed on the silicon substrate. With electrons is 0, without electrons is 1.

Just like its name, the flash memory is initialized by deleting data before writing. Specifically, electrons are derived from all floating gates. Some data will be returned to "1".

When writing, only write when the data is 0, and do nothing when the data is 1. When writing 0, a high voltage is applied to the gate electrode and the drain, increasing the energy of electrons conducted between the source and the drain. As a result, electrons will break through the oxide film insulator and enter the floating gate.

When reading data, a certain voltage is applied to the gate electrode, the current is large, and the small current is set to 0. In a state where the floating gate has no electrons (data is 1), when the voltage is applied to the gate electrode, a voltage is applied to the drain, and a large amount of electrons move between the source and the drain to generate a current. In the floating gate with electrons (data is 0), the conduction electrons in the channel will decrease. Because the voltage applied to the gate electrode is absorbed by the floating gate electrons, it is difficult to affect the channel.

Application prospect

"USB flash drive" is the most obvious portrayal of flash memory in daily life. In fact, flash memory has appeared in many electronic products before the U disk. The traditional way of storing data is the volatile storage of RAM, the data will be lost when the battery runs out

Flash Card

Flash Card

Lost. Products using flash memory overcome this problem and make data storage more reliable. In addition to flash disks, flash memory is also used in electronic products such as BIOS, PDAs, digital cameras, voice recorders, mobile phones, digital TVs, and game consoles in computers.

Back in 1998, USB flash drives entered the market. The interface has developed from USB1.0 to 2.0 to the latest USB3.0, and the speed has gradually increased. The prevalence of U disk also indirectly promoted the promotion of USB interface. Why is the U disk so popular with people?

Flash drives can be used to exchange data between computers. In terms of capacity, the capacity of the flash disk is selectable from 16MB to 64GB, breaking the limitation of 1.44MB of floppy drive. In terms of reading and writing speed, the flash disk uses a USB interface, and the reading and writing speed is much higher than that of a floppy disk. In terms of stability, the flash disk does not have a mechanical reading and writing device, which avoids damage caused by the mobile hard disk that is easy to be injured or dropped. Some styles of flash drives have encryption and other functions, which make users more personalized. The flash drive is small and easy to carry. And it adopts a USB interface that supports hot swap, which is very convenient to use.

Flash memory is developing in the direction of large capacity, low power consumption, and low cost. Compared with traditional hard disks, flash memory has high read and write speeds and low power consumption. Flash memory hard disks, namely SSD hard disks, have appeared on the market. With the improvement of manufacturing process and the reduction of cost, flash memory will appear more in daily life.

Determining factors

Number of pages

As mentioned earlier, the larger the capacity of the flash memory, the more pages and pages, the longer the addressing time. But the extension of this time is not a linear relationship, but a step by step change. For example, chips of 128 and 256Mb require 3 cycles to transmit address signals, 512Mb and 1Gb require 4 cycles, and 2 and 4Gb require 5 cycles.

Page capacity

The capacity of each page determines the amount of data that can be transferred at one time, so large-capacity pages have better performance. As mentioned earlier, large-capacity flash memory (4Gb) increases the page capacity from 512 bytes to 2KB. The increase in page capacity is not only easy to increase capacity, but also improves transmission performance. We can give an example. Take Samsung K9K1G08U0M and K9K4G08U0M as examples, the former is 1Gb, 512 byte page capacity, random read (stable) time 12μs, write time is 200μs; the latter is 4Gb, 2KB page capacity, random read (stable) time 25μs, write time 300μs. Suppose they work at 20MHz.

Read performance

The reading steps of NAND flash memory are divided into: sending commands and addressing information → transferring data to the page register (random read stabilization time) → data transfer (8bit per cycle, need to transfer 512 + 16 or 2K + 64 times).

K9K1G08U0M needs to read one page: 5 commands, addressing period×50ns+12μs+(512+16)×50ns=38.7μs; K9K1G08U0M actual read transfer rate: 512 bytes÷38.7μs=13.2MB/s; K9K4G08U0M read one page Need: 6 commands, addressing cycle × 50ns + 25μs + (2K + 64) × 50ns = 131.1μs; K9K4G08U0M actual read transfer rate: 2KB bytes ÷ 131.1μs = 15.6MB/s. Therefore, using 2KB page capacity to improve read performance by about 20% than 512-byte page capacity.

Write performance

The writing steps of the NAND flash memory are divided into: sending addressing information → transferring data to the page register → sending command information → writing data from the register to the page. The command cycle is also one. We combine it with the addressing cycle below, but these two parts are not continuous.

K9K1G08U0M needs to write one page: 5 commands, addressing period×50ns+(512+16)×50ns+200μs=226.7μs. K9K1G08U0M actual write transfer rate: 512 bytes÷226.7μs=2.2MB/s. K9K4G08U0M needs to write a page: 6 commands, addressing period × 50ns + (2K + 64) × 50ns + 300μs = 405.9μs. K9K4G08U0M actual write transfer rate: 2112 bytes/405.9μs=5MB/s. Therefore, the use of 2KB page capacity than 512 byte page capacity to improve write performance more than twice.

Block capacity

The block is the basic unit of the erasing operation. Since the erasing time of each block is almost the same (the erasing operation generally takes 2ms, and the time occupied by the command and address information in the previous cycles is negligible), the capacity of the block will directly determine Erase performance. The page capacity of the large-capacity NAND flash memory has increased, and the number of pages per block has also increased. Generally, the block capacity of a 4Gb chip is 2KB×64 pages=128KB, and the size of a 1Gb chip is 512 bytes×32 pages=16KB. It can be seen that within the same time, the wiping speed of the former is 8 times that of the latter!

I/O bit width

In the past, there were generally 8 data lines for NAND flash memory, but from 256Mb products

At the beginning, products with 16 data lines appeared. However, due to controllers and other reasons, the actual application of x16 chips is relatively small, but the number will still show an upward trend in the future. Although the x16 chip still uses a group of 8 bits when transmitting data and address information, and the occupied period is also the same, but when transmitting data, it takes 16 bits as a group, and the bandwidth is doubled. K9K4G16U0M is a typical 64M×16 chip. It is still 2KB per page, but the structure is (1K+32)×16bit.

Imitating the calculation above, we get the following. K9K4G16U0M needs to read one page: 6 commands, addressing period×50ns+25μs+(1K+32)×50ns=78.1μs. K9K4G16U0M actual read transfer rate: 2KB bytes ÷ 78.1μs = 26.2MB/s. K9K4G16U0M needs to write a page: 6 commands, addressing cycle × 50ns + (1K + 32) × 50ns + 300μs = 353.1μs. K9K4G16U0M actual write transfer rate: 2KB bytes÷353.1μs=5.8MB/s

It can be seen that with the same capacity chip, after the data line is increased to 16, the read performance is improved by nearly 70%, and the write performance is also improved by 16%.

Frequency

The effect of operating frequency is easy to understand. The working frequency of NAND flash memory is 20-33MHz, the higher the frequency, the better the performance. Taking K9K4G08U0M as an example, we assume that the frequency is 20MHz. If we double the frequency to reach 40MHz, then K9K4G08U0M needs to read a page: 6 commands, addressing period × 25ns + 25μs + (2K + 64) × 25ns = 78μs. K9K4G08U0M actual read transfer rate: 2KB bytes÷78μs=26.3MB/s. It can be seen that if the operating frequency of K9K4G08U0M is increased from 20MHz to 40MHz, the read performance can be improved by nearly 70%! Of course, the above example is just to facilitate the calculation. In Samsung's actual product line, it is K9XXG08UXM that can work at a higher frequency, not K9XXG08U0M, the former frequency can reach 33MHz.

Manufacturing process

The manufacturing process can affect the density of the transistor and also the time of some operations. For example, the aforementioned write stabilization and read stabilization time, they occupy an important part of time in our calculations, especially when writing. If you can reduce these times, you can further improve performance. Can the 90nm manufacturing process improve performance? The answer is probably no! The actual situation is that with the increase in storage density, the required read and write stabilization time is showing an upward trend. The example given in the previous calculation reflects this trend, otherwise the performance improvement of the 4Gb chip will be more obvious.

Overall, although the addressing and operation time of large-capacity NAND flash memory chips will be slightly longer, as the page capacity increases, the effective transfer rate will still be larger. Large-capacity chips meet the market's requirements for capacity, cost, and performance. Demand trends. Increasing the data line and increasing the frequency are the most effective ways to improve performance. However, due to the influence of process and physical factors such as command and address information occupying the operation cycle and some fixed operation time (such as signal stabilization time, etc.), they will not Bring year-on-year performance improvement.

1Page=(2K+64)Bytes; 1Block=(2K+64)B×64Pages=(128K+4K)Bytes; 1Device=(2K+64)B×64Pages×4096Blocks=4224Mbits

Among them: A0 ~ 11 address the page, can be understood as "column address".

A12 to 29 address pages, which can be understood as "row addresses". For convenience, the "column address" and "row address" are divided into two groups for transmission, rather than directly combining them into a large group. Therefore, each group will have several data lines without information transmission in the last cycle. Unused data lines remain low. The so-called “row address” and “column address” of NAND flash memory are not the familiar definitions in DRAM and SRAM, but just a relatively convenient expression. For ease of understanding, we can make a vertical section of the above three-dimensional NAND flash memory chip architecture diagram, and apply the two-dimensional "row" and "column" concepts in this section to be more intuitive.

Development process

Development History

In 1984, the inventor of Toshiba Co., Ltd. Fujioka first proposed the concept of flash memory (referred to here as flash memory). Unlike traditional computer memory, flash memory is characterized by non-volatility (that is, the stored data will not be lost after the host is powered off), and its recording speed is also very fast.

Intel is the first company in the world to produce flash memory and put it on the market. In 1988, the company introduced a 256K bit flash memory chip. It is the size of a shoe box and is embedded in a recorder. Later, this type of flash memory invented by Intel was collectively called NOR flash memory. It combines EPROM (Erasable Programmable Read Only Memory) and EEPROM (Electrically Erasable Programmable Read Only Memory) technologies and has an SRAM interface.

The second type of flash memory is called NAND flash memory. It was developed by Hitachi in 1989 and is considered an ideal replacement for NOR flash memory. The write cycle of NAND flash memory is 90% shorter than that of NOR flash memory, and its save and delete processing speed is relatively fast. The storage unit of NAND is only half of NOR, and NAND obtains better performance in a smaller storage space. In view of the excellent performance of NAND, it is often used in memory cards such as CompactFlash, SmartMedia, SD, MMC, xD, and PC cards, USB sticks, etc.

Market analysis

The flash memory market is still in an immature period when the heroes are fighting for hegemony. Samsung, Hitachi, Spansion and Intel are the four major manufacturers in this market.

Due to some strategic errors, Intel gave up its top seat for the first time and fell behind Samsung, Hitachi and Spansion.

AMD Flash memory business unit Spansion produces both NAND and NOR flash memory. Its NOR flash memory output in the first half of the year was almost the same as that of Intel, becoming the largest manufacturer of NOR flash memory. The company's profit in the first half of the year was $1.3 billion, which is almost more than half of its entire company's profit ($2.5 billion).

Overall, Intel and AMD achieved gratifying results in the first half, but Samsung and Hitachi suffered setbacks.

According to estimates by market research firm iSuppli, global flash memory revenue will reach 16.6 billion US dollars, up 46% from 2003 (11.64 billion US dollars). The source's demand for digital cameras, USB sticks and compressed MP3 player memory will greatly drive the sale of flash memory. It is predicted that flash memory sales will reach 17.5 billion US dollars in 2005. However, iSuppli estimates that the profit increase of flash memory from 2005 to 2008 will fall back to a maximum of 22.4 billion US dollars.

Alternatives

Compared with many short-lived information technologies, flash memory has fully demonstrated its "older generation" style with its 16-year history. In the early 1990s, flash memory first entered the market; by 2000, the amount of profits had exceeded one billion US dollars. Peter, director of Infineon’s flash memory department, once said, “As far as the life cycle of flash memory is concerned, we are still in a rising stage.” Infineon believes that the sales of flash memory still have room to rise and is preparing to join the market Investment. Infineon announced that its 200mm DRAM plant in Dresden has started to produce 512Mb NAND compatible flash memory chips. By the end of 2004, Infineon plans to use a 170-nanometer manufacturing process to manufacture more than 10,000 wafers per month. In 2007, the company hoped to become the top three in the NAND market.

In addition, Stefan Lai, deputy general manager of Intel Technology and Manufacturing Group, believes that before 2008, flash memory will be irreplaceable. In 2006, Intel will first adopt 65-nanometer technology; by 2008, the next-generation 45-nanometer technology under development will be expected to be put on the market. Stefan Lai feels that the prediction is still relatively shallow, perhaps 32 nanometers, 22 nanometers technology is entirely possible. But Stefan Lai also admitted that from 2008 to 2010, new technology may replace it.

Despite the increasing incentives for the discussion of flash memory replacements, flash memory is still valued by the market. Future alternatives must not only be non-volatile memory like flash memory, but also slightly better in speed and write cycle. In addition, production costs should also be relatively low. Since the manufacturing technology is not yet mature, new alternatives will not pose an absolute threat to flash memory.

Compared with hard disk

If only from the storage medium, flash memory is better than hard disk. This refers to the speed of data transmission and the degree of shock resistance (there is no shock resistance in flash memory).

advantage:

1. The size of the flash memory is small. This does not mean that the integration of flash memory will be high. The main reason why the micro hard disk is so large is that the micro hard disk cannot be made smaller than the flash memory, and it does not mean that the micro hard disk is not highly integrated.

2. Compared with the hard disk, the flash memory structure is not afraid of shock and is more resistant to falling. The hardest thing about the hard disk is the strong vibration. Although we can be very careful when we use it, but tigers also doze off.

3. Flash memory can provide faster data reading speed, while the hard disk is limited by the speed.

4. Flash storage of data is more secure, for reasons including:

1. Its non-mechanical structure, so movement will not affect its reading and writing;

2. The service life of widely used mechanical hard drives is very much related to the number of reads and writes and the speed of reading and writing, but the flash memory is not greatly affected;

3. The writing of the hard disk is written by magnetism, the flash memory uses voltage, and the data will not be erased due to time.

5. The quality is lighter.

Disadvantages:

1. The material is expensive, so the unit capacity is more expensive.

2. The reading and writing speed is relatively slow.