Multi-Level Cell Memory Seeks Enterprise SSD Status

Price continues to be a major sticking point when it comes to the enterprise adoption of solid-state drives (SSDs), so it is not surprising that multi-level cell memory (MLC) continues to surface in industry discussions and product launches concerning SSDs. MLC costs significantly less than its single-level cell memory (SLC) counterpart.

MLC carries four bits instead of SLC's two bits. As a result, in addition to the lower price, it delivers twice the capacity of SLC and the extra capacity improves MLC memory life cycle over that of SLC. Unfortunately, MLC's multiple logic levels complicate the write process to the point where MLC delivers write endurance results in the thousands, as compared to a write endurance rating that is in the millions for SLC. MLC has a much lower write endurance rating because a complex of multiple logic levels must interact with each other, and the interaction contributes to earlier wear out than SLC. Just as significantly, MLC memory has traditionally failed sooner than SLC memory due to MLC's smaller voltage tolerance.

However, this does not prevent companies and industry analysts from projecting a role for MLC in the enterprise. "You have to believe that as the technology progresses, MLC will become capable and reliable enough for the enterprise market," says Arun Taneja, founder of industry researcher Taneja Group. He expects that MLC will assume a position in the low end of the enterprise market.

Part of this enterprise "low end" market consists of drives for notebook computers, which ST Microelectronics began shipping in 2007 and Samsung and others began shipping in 2008.

"There are still challenges in MLC performance and error rates," says Khandker Quader, senior vice president of memory technology and product development for SanDisk, which recently announced a 4-bits-per-cell (X4) MLC memory chip. SanDisk, which considers itself a vertical integration company, has pursued an approach that combines an X4 controller chip with an MLC memory chip in a multi-chip package. "The controller is a critical part of the product," says Quader. "It manages the memory for better performance and it has an innovative error handling scheme that allows us to address some of the shortcomings of MLC."In SanDisk's X4 MLC deployment, the proprietary controller uses SanDisk-originated algorithms that allow more bits within MLC memory to experience failure without memory becoming unavailable. "One of the functions a controller addresses is making sure that wear is evenly distributed across memory cells," says Taneja. "So if 95 percent of your memory is empty and 5 percent is occupied -- and if you delete some of the data in that 5 percent of memory that is occupied -- the new data that is written to memory is written to cells that have not been used before, and the 5 percent of memory that you used in the past is not used. This spreads the wear of memory cells evenly, and it is the controller that does the tracking."

SanDisk says its MLC controllers and memory have in some cases placed MLC on par with the performance of SLC, but it also says that MLC at this time is best deployed in consumer applications like memory sticks and digital cameras. "We are looking to leverage MLC into the enterprise market, but we are cautiously optimistic," says Quader.

One enterprise barrier to MLC adoption is the mean time between failure. Another remains the number of write cycles that MLC can handle before it fails. "An enterprise will not accept product that produces fewer than 1 million hours between failures, and the number of writes that enterprises require are in the hundreds of millions and maybe more," Taneja says.

Still, the trends seem favorable for improvement in MLC technology. Innovations in MLC, especially in silicon geographies that continue to shrink, will continue to exert downward price pressures on both MLC and SLC memory. Intellectual property and proprietary controller and chip packaging architectures and strategies also will continue to play a role in MLC, and will need to be carefully evaluated by enterprises in their purchase processes. A good example is the recent Toshiba-SanDisk partnership, where the manufacturing and development of MLC memory is shared, but where each company develops and integrates its own controller with the resulting memory.

Eventually, all paths for MLC will lead to the enterprise. While some MLC purveyors are already in the enterprise market, many more that are comfortably entrenched in the consumer market hope to leverage their product so that it ultimately meets enterprise criteria. This will inspire healthy competition.0