Process technology has allowed significant density increases and lower cost-per-Mbyte since flash memory was first introduced in 1988. Intelís first flash device was a 256Kb device priced at $20 ($640 per Mbyte). Since then process innovation has increased the density and decreased the cost-per-Mbyte. In 1996, for example, an 8 Mbit device was priced at $12 ($12 per Mbyte) - more than a 50X reduction in cost-per-Mbyte. Continued process scaling will inevitably drive up density and lower cost-per-Mbyte. To build on process scaling, a concept called MultiLevel Cell (MLC) technology was conceived. This involves storing multiple bits of information on a single memory transistor. Storing two bits per memory cell instantly doubles the density in the same space and lowers the cost-per-Mbyte. MLC, together with process scaling, is poised to hit $1 per Mbyte by 2001.

The expanding need for additional non-volatile code or data storage by applications creates demand for higher and higher density flash products. In addition, these demands also include decreases in the cost-per-Mbyte. Typical process scaling has taken flash from a 256Kb device at $640 per Mbyte to an 8Mb device at $12 per Mbyte in 8 years. Continuous innovations will continue this density increase with lower cost-per-Mbyte. MultiLevel Cell technology intensifies this learning by adding multiple bits of information on a single memory transistor. This significantly increases the number of bits stored per area while effectively decreasing the overall cost-per-megabyte.

A NOR-based Multi-Level Cell architecture provides direct memory-cell access which is essential in providing accurate charge sensing, placement, and storage --three key aspects of MLC technology. This direct memory cell access results in a reliable, robust technology suitable for many applications.

This article describes the fundamentals of a flash MLC technology, providing insight to device operating characteristics of programming and reading. In addition, it briefly describes system implementation advantages associated with the MLC technology path.

Standard Flash Technology
Typical flash memory uses a single bit-per-cell. Each cell is characterized by a specific threshold voltage, or V t level. Electrical charge is stored on the floating gate of each cell. Within each cell, or transistor, two possible voltage levels exist (Figure 1, a). These two levels are controlled by the amount of charge that is programmed or stored on the floating gate; if the amount of charge on the floating gate is above a certain reference level, the cell is considered to be in a different level.

MLC Technology
Multi-Level Cell (MLC) technology enables storage of multiple bits per memory cell by charging the polysilicon floating gate of a transistor to different levels. This technology takes advantage of the analog nature of a traditional flash cell by assigning a bit pattern to a specific voltage range. MLC effectively reduces cell area as well as the die size for a given density. This ultimately leads to a significantly reduced unit cost-per-megabyte --one of the greatest benefits of this technology. Todayís MLC memory products are capable of storing two bits per memory cell (Figure 1, b). The MLC technology path illustrates that NOR-based MLC technology is easily scaleable to three bits-per-cell (Figure 1, c), which enables another significant increase in density per area.

In MLC technology, the voltage across each cell has been divided into greater than two levels. Figure 1 illustrates Vt level placement for two and three bit-per-cell technology.

MLC Cell Operations
MLC devices must be able to manage electrical charge precisely. More specifically, these devices must be able to place charge with precision, sense charge with precision, and store charge over time. These requirements are met by the NOR architectureís direct connection to the gate, source, and drain of each memory cell. Each cell is made up of a single transistor which is directly connected to the appropriate control voltages in order to accomplish pinpoint accuracy in charge manipulation.

MLC Program Operation
During programming, a precise charge must be placed onto the floating gate. The analog voltage that exists across each flash cell is divided into multiple Vt levels which are controlled by the amount of electrons on the floating gate. Channel Hot-Electron (CHE) injection controls this charge placement.

During programming, each cellís direct ground, bitline, and wordline connection enables precise charge placement. A cellís control gate links to the internally generated supply voltage through a direct wordline connection and row decoding. The drain is pulsed at a constant voltage through a direct bitline connection and column decoding. The source is directly connected to ground (Figure 2, a).

Electron storage on the floating gate creates a potential that must be overcome by the control gate. This potential results in a higher turn-on threshold voltage (Vt for that transistor (memory cell). Direct, precise gate and drain control is critical to MLC memory cell placement.

In MLC technology, the voltage across each cell has been divided into multiple levels. Figure 4 illustrates cell distribution over individual voltage levels, along with the placement reference points (P1, P2, and P3) for two bit-per-cell technology. During programming, a specific voltage range is used to represent one of the specific levels. Programming data at a given memory address transitions selected cells from the "11" (erased) level to the "10," "01," or "00" (programmed) levels. Storing two bits-per-cell requires four levels defined by read reference cells; R1, R2, and R3. These voltage levels are defined by the threshold voltage of the read reference cells as follows: Vt <; Vt(R1) is level 0, Vt(R1) <; Vt <; Vt(R2) is level 1, Vt(R2) <; Vt <; Vt(R3) is level 2, and Vt >; Vt(R3) is level 3. Specific, unique voltage values define each level.

MLC Read Operation
During data read, charge sensing is of maximum importance. Through direct connections to each memory cell, the data read operation determines the level of each memory cell quickly, accurately, and reliably. Read speeds are comparable to single bit-per-cell technology.

The data read operation senses which of the four levels the memory cell falls within based on the threshold voltages of the three read reference cells. The reference cells are biased in such a way that each conducts a current proportional to the Vt of its specific level. During the read operation; VREAD is placed on the control gate, the source is grounded, and a drain bias is applied. In this mode, a cell conducts a current (ICell) which is proportional to the cellís Vt.

When the cells are biased in this way, the current conducted through each cell has an inverse relationship to its Vt. Therefore, if current through the flash array cell is greater than the appropriate reference current (ICell >; I R#Ref ) then the memory cell Vt is less then the reference cell Vt (VtCell<; VtRef). The data is read by comparing the current through the memory cell to the current through three read reference cells.

The bitline current for the memory cell is compared to the current produced by three read reference circuits. The current is sensed by connecting the drain to an active load, which is connected to a differential sense amplifier. For a read operation, there is a sense amplifier for each of the three read reference cells. Each sense amplifier has a flash array input, from the bitline, and a reference cell input. If I Cell >; IR# the sense amplifier outputs a logic "1."

The outputs of the three sense amplifiers are connected to a logic circuit which interprets the two data bits in parallel. If the memory cell is in level 0 then I Cell >; I R1Ref. Therefore, all of the sense amplifiers output a logic "1" and the logic circuit outputs the data "11." In level 1, I R2Ref <; I Cell <; I R1Ref, so sense amplifier 1 outputs a logic "0" and sense amplifiers 2 and 3 output a logic "1" The logic circuit interprets these outputs to be data "10." In level 2, I R3Ref <; I Cell <; I R2Ref, so sense amplifiers 1 and 2 output a logic "0" and sense amplifier 3 outputs a logic "1." The logic circuit interprets these outputs to be data "01." Finally in level 3, I Cell <; I R2Ref, so all the sense amplifiers output a logic "0" and the logic circuit outputs the data "00."

The sensing operation described here for the data read operation is also used in the program operation. During the program operation, sensing is done to ensure that the memory cell has reached the desired level.

MLC Technology Advantages
There are many advantages associated with the use of flash memory. The solid-state nature of flash provides ruggedness far superior to mechanical rotating media. Not only is MLC memory valuable as a digital media, but advantages such as direct access to each cell, specific to NOR-based flash architectures, permit reliable charge placement, sensing, and storage.

Lower Cost
Cost relates directly to die size. The smaller the die size, the lower the cost. Dramatic cost breaks are associated with the MLC technology approach because, at two bits-per-cell, the density of the device is doubled without increasing the die size. This becomes even more apparent at three bits-per-cell, where the density of bits-per-cell will triple without significantly increasing die size!

Other Flash Memory Technology Advantages

Several NOR-based flash vendors offer Common Flash Interface (CFI), making compatibility among future flash devices. CFI is analogous to putting a "datasheet on a chip." Software that takes advantage of CFI devices simply poll CFIís query ROM for parameters such as device size, buffer size, and command set support. The key CFI benefit is the ease of upgrading.

MLC technology allows storage of multiple bits per memory cell by charging memory cell to different voltage levels. These advances significantly increase the number of bits stored per area and decrease the cost-per-megabyte. At two bits-per-cell, densities double, and at three bits-per-cell densities triple; both cases yield a significant increase in density without a significant increase in die size.

NOR-based MLC technology uses simple, fundamental techniques and algorithms for program, read, and erase. The direct memory cell access of NOR-based MLC is essential in providing accurate charge sensing, placement, and storage--three critical aspects of MLC technology. This approach results in reliable, robust flash memory suitable for many applications.

MLC memory products will cost-effectively satisfy many different applications, such as memory cards, solid-state drives, resident code and file storage, data acquisition, and embedded code storage. Applications that take advantage of Common Flash Interface allow for compatibility among future flash devices.