e.Digital Corporation (fka EDIG)

[culater33]

Match the flash memory to the application

Richard Nass, Editor-in-Chief

Different types of flash memory are available, ranging from multiple chip offerings to multiple subassemblies. It's important to pick the one that matches your need.

Flash is flash, right? The answer to that is yes and no. In general, there are two types of flash memory—NAND and NOR. NAND-based flash memory is generally used for data storage, while NOR-based memory is used for code storage. NAND's sequential (serial) access suits it for block-oriented data storage applications.

Portable devices are the perfect location for flash memory, particularly when compared with rotating media (except for the cost). In general, they're smaller and lighter than rotating media, two key attributes of portable electronics.

One trend that's occurring in the market is that designers are looking at NAND flash as an alternative to NOR, even for code-storage applications. The reason for this is purely due to cost. On a per-bit basis, NOR is two to three times more expensive than NAND.

Click here to enlarge image

Fig. 1. SST's SuperFlash technology employs a split-gate cell, in which the erase oxide is neither the gate nor the transistor oxide.

The amount of storage that's needed in cell phones continues to rise. The number of features designed into the latest phones is becoming more complex, driving that increase. Also, simple features like upping the available phone-book size or memory retention adds to the memory requirement.

But the most dramatic leap in memory footprint will come when the transition to 3G phones occurs. I'll save the guess for when that will occur for another article. But we can safely assume it will be within the next five years. Such phones will integrate an MP3 player, an Internet browser, a color display, or a camera.

The same transition should occur in the PDA arena, systems that don't want to employ rotating media. The NAND flash will be used to store everything, including the operating system, the application software, and the user data storage.

Some vendors have a vested interest in which type of memory will be the most dominant. Then there are vendors like Toshiba who produce both types. Its NAND offerings range from 64 Mbits to 1 Gbit, while the NOR devices are available at 16, 32, and 64 Mbits.

Souped-up NOR
Silicon Storage Technology (SST) offers a NOR-type flash using its SuperFlash split-gate cell architecture (Fig. 1). The technology uses a reliable thick-oxide process and a simple, flexible design that's suitable for small or medium sector sizes.

With SuperFlash technology, erasing is accomplished using a thick polysilicon oxide as opposed to gate oxide. As a result, the memory devices easily scale to finer process geometries. Programming is achieved by a low-power technique called source-side injection.

The latest developments in SuperFlash yield a self-aligned cell, which removes the need for built-in tolerances on the various process layers (Fig. 2). Most foundries will be able to align layers perfectly to each other without the need to build tolerances into those layers.


Fig. 2. A self-aligned cell is now part of the SuperFlash technology. This helps achieve smaller devices by removing the need for built-in tolerances on the various process layers.
Even within the NAND camp, a diversified portfolio can be helpful. This could include different package types (generally BGA or chip-scale packages) and lower voltages. Toshiba is now readying a 1.8-V offering.

According to Kevin Kilbuck, director of business development for flash memory at Toshiba, "We're addressing the lower voltages in two steps. One is to take our existing 3-V products and take them down to 2.7 V. Step two is a pure 1.8-V core. There's not a huge demand for that today, but as the 3G phones start to roll and as PDAs move to NAND, our timing is basically matched to those markets."

Another consideration in the NAND versus NOR discussion has to do with the controller. Today, most systems have some type of controller between the host processor and the NAND. The interface to NOR flash is similar to SRAM, something that designers are generally familiar with. Hence, the flash can be interfaced directly to the microprocessor bus. On the flip side, next-generation NAND flash will be designed to permit that direct processor connection without any glue logic.

Advanced Micro Devices (AMD) is one of the vendors that currently offers 1.8-V flash memory (operating up to 2.2 V). The Am29SL and Am29DS families offer single power-supply operation, as well as zero-power operation and advanced power management.

The Am29DS devices offer a dual-bank, simultaneous read/write capability, while the Am29SL family features a standard single-bank architecture. The zero-power operation reduces typical current draw to as low as 1 µA in standby and automatic sleep modes. Eliminating the need for higher external programming and erasing voltages reduce overall system cost and conserve board space.

Intel's 1.8-V flash memory integrates a flexible partition read-while-write architecture with synchronous burst and asynchronous page-mode read operations. This is combined with the security-enabling features of the company's Advanced+ Boot Block. The memory is supported by Intel's Flash Data Integrator software (Version 4), which enables management of code, data, and files in flash memory. Densities of 32, 64, and 128 Mbytes are available, in various packaging options.

STMicroelectronics recently announced a 32-Mbit flash chip, the M58LW032A, that can store data, but its high performance also allows for direct execution of stored code, precluding the need for a separate RAM. The device performs all operations, including programming and erasing, from a 2.7- to 3.6-V supply, and accepts I/O signals ranging from 1.8 V to the supply voltage.

Two members of the Firmware Hub (FWH) family of flash memories hail from Atmel. The AT49LW040 is a 4-Mbit device, while the AT49LW080 stores 8 Mbits. The BIOS-based components can interface directly with Intel's 8xx series chip sets. The parts feature 64-kbyte sectors and automated byte-program and sector-erase operations. Both parts are available in 40-lead TSOP and 32-lead PLCC surface-mount packages.

Software control
A software product called VBM can be coupled with NAND memory to help manage bad blocks. Developed by Datalight, VBM is bundled with the FlashFX flash media-management software. FlashFX is portable across most operating systems and flash devices. Hence, it doesn't lock a system manufacturer into any particular flash vendor. Version 5.0 is now available, a release that's targeted at OEMs building Windows CE-based devices

Going in somewhat of a different direction, Matrix builds memory devices that are low cost, yet can only be written to once. The low cost comes from the ability to produce the memory in a three-dimensional format. The first devices to come from this format, available in the second half of this year, will be a 64-Mbyte memory chip in a standard TSOP package.

"Until now, anytime you've built an IC, you basically built the chip by creating transistors on the surface of a single-crystal silicon wafer. Then, you wire them together," says Dan Steere, vice-president of marketing at Matrix. "The result is a memory that's denser than existing memory, and therefore cheaper."

What makes this technique even more valuable is that it can be achieved using standard materials and standard processes, meaning that the wafers are produced with the equipment and materials that are already in most high-volume CMOS fabs.

The initial target for the write-once memory is digital cameras. The company says that with a 3-Mpixel camera, users can store about 75 pictures. With a 1-Mpixel camera, that increases to between 250 and 300 pictures. And the retention life is rated at 100 years. The price goal for the memory cards (which is set by Matrix's customers) is to be about the same as a three-pack of 35-mm film.

While this creates a different category of memory, the devices are designed to be compatible with existing memory-card standards, including a standard NAND flash interface, so the cards can be plugged into the same slots.

Flash combo
The spin from Microchip is to combine a microcontroller with the flash memory. For example, the company's PIC18F6720 and PIC18F8720 devices offer a 1-Mbit flash array that can be fully erased and programmed in less than 2 seconds. An individual word can be erased and programmed in less than 3 ms.

The parts' feature set includes an analog-to-digital converter with up to 16 10-bit channels, and up to 10-MIPS performance at 40 MHz. The operating voltage ranges from 2 to 5.5 V.

Micron's SyncFlash memory is designed such that both DRAM and flash memory can reside on the same bus and execute from the same DRAM memory controller. This simplifies system busing by eliminating the additional pins needed for a separate flash-only memory interface. It can also increase flash read performance to DRAM speeds. The first member of the SyncFlash product family is a 64-Mbit device, housed in a 54-pin TSOP Type II package.

The HY29DS32x, developed by Hynix, is a 32-Mbit, 2-V flash memory available in 48-pin TSOP and 48-ball FBGA packages. The memory array is organized into 71 sectors in two banks. The first bank contains eight 8-kbyte boot/parameter sectors and 7 or 15 larger sectors of 64 kbytes each (depending on the version of the device). The second bank contains the rest of the memory array, organized as 56 or 48 sectors of 64 kbytes.

The device features simultaneous read/write operation, with zero latency. This releases the system from waiting for the completion of program or erase operations, improving overall system performance. After a program or erase cycle has been completed, the device is ready to read data or to accept another command. Reading data out of the device is similar to reading from other flash or EPROM devices.

Samsung Electronics offers a 1-Gbit NAND flash chip, which is fabricated with 0.12-micron technology. The device is offered in a single chip as well as a dual-die package. The dual-die package doubles the capacity to 2 Gbits. The initial 1-Gbit device operates at 3.3 V. A 1.8-V version will be available shortly.

The chip is built with an expanded 2-kbyte page program (rather than the standard 512 bytes), while the block erase is 128 kbytes (unlike the standard 16 kbytes), improving write performance. A write-cache function is invoked when continuous page programming is performed.

SanDisk is a vendor that produces flash memory in a host of form factors, including CompactFlash, SmartMedia, and MutliMedia Card. In addition, they've developed a 1-Gbit NAND flash chip, the SDTNF-1024. The part operates at 3.3 V and is organized as 528 bytes by 32 pages by 8192 blocks. The 528-byte static register allows program and read data to be transferred between the register and the memory cell array in 528-byte increments. Erase operations are implemented in a single block unit.

Instead of hard disk drives
Simpletech supplies a range of flash readers that lets users access the flash memory.

"We're seeing a real need in the industrial space, where people tend to use these flash devices as alternatives to hard disk drives," says Dan Moczarny, director of flash and storage products at Simpletech. "The flash devices run on the order of 0.6 W, versus the traditional 2.5- or 3.5-in. hard drives, which are in the 3- to 8-W range. We can also operate from -40°C to +85°C."


Fig. 3. Multiple vendors, including SimpleTech, are offering flash memory in a CompactFlash form factor.
Simpletech has an interesting perspective on pricing. The company claims that if the designer only needs around 100 Mbytes of storage, then flash can be less expensive than traditional rotating media.

Says Moczarny, "We look at price per usable megabyte, rather than price per megabyte. Right now, if you look at ATA disk drives, minimum capacities in volume are in the 6-Gbyte range. There are many systems using embedded Linux or Windows CE, where the storage requirements are under 100 Mbytes. We see a price-crossover point where we could probably get into the 150- to 200-Mbyte range on a flash product for about the same price as a 6- to 10-Gbyte hard drive."

Simpletech recently announced a new line of Secure Digital (SD) memory cards in capacities up to 128 Mbytes, aimed at PDAs, cell phones, and digital cameras. Higher densities are expected soon. The company also offers a line of CompactFlash cards (Fig. 3).

SD cards offer built-in copy protection to ease fast, secure downloading of digital files, with a typical transfer rate of 2 Mbytes/s. A 128-Mbyte SD card will hold over 40 minutes of video, 100 digital images, or up to 4 hours of music.

Smart Modular Technologies specializes in high-speed flash devices. According to Steffen Hellmold, director of the company's flash group, "We'll exceed the 5-Mbyte/s mark by the end of this year. We'll do that through enhancements in the controller architecture, as well as enhancements in the flash ICs."

Smart currently offers a 1-Gbyte CompactFlash card in a Type II format. It can operate at either 3 or 5 V.

Advanced Micro Devices
Sunnyvale, CA
[phone]
www.amd.com

Atmel
San Jose, CA
[phone]
www.atmel.com/atmel/acrobat/doc1966.pdf

Datalight
Bothell, WA
[phone]
www.datalight.com

Hynix Semiconductor America
San Jose, CA
[phone]
www.us.hynix.com

Intel
Santa Clara, CA
[phone]
developer.intel.com/design/flash/

Matrix Semiconductor
Santa Clara, CA
[phone]
www.matrixsemi.com

Microchip Technology
Chandler, AZ
[phone]
www.microchip.com

Micron Technology
Boise, ID
[phone] or [phone]
www.micron.com

Samsung Electronics
San Jose, CA
408-544-4000
www.samsungusa.com

Silicon Storage Technology (SST)
Sunnyvale, CA
[phone]
www.sst.com

SimpleTech
Santa Ana, CA
[phone] or [phone]
www.simpletech.com

STMicroelectronics
Lexington, MA
[phone]
www.st.com

Toshiba America Electronic Components
Irvine, CA
[phone]
www.toshiba.com/taec

Portable Design April, 2002
Author(s) : Richard Nass
http://pd.pennnet.com/Articles/Article_Display.cfm?Section=Articles&Subsection=Display&ARTIC...
culater