Search the Community
Showing results for tags 'next-gen'.
IBM has demonstrated a new type of memory technology that the company believes could one day be a replacement for NAND flash. The company’s Theseus Project (conducted in cooperation with the University of Patras in Greece) is the first attempt to combine phase change memory, conventional NAND, and DRAM on a single controller. The result? A hybridized storage solution that outperforms PCIe-based SSDs by between 12 and 275 times. The physics of phase change Phase change memory is one of a number of alternative memory structures that’s beenproposed as a replacement for NAND. Phase change memory works by rapidly heating chalcogenide glass, shifting it between its crystalline and amorphous state. In its amorphous state (read as a binary 0), the structure has very high resistance, while in its crystalline state (binary 1) resistance is quite low. Phase change memory can quickly shift between the two states, plus research from Intel and Micron have demonstrated the feasibility of intermediate states, which allows two bits of information to be stored per cell. Phase change memory has much lower latency than NAND, much faster read/write times (in theory), and it can withstand millions of write cycles as compared to 30,000 with high-end SLC NAND and as few as 1,000 with TLC NAND. Even better, it’s well positioned compared to other theoretical memory devices. Even so, NAND flash has enormous economies of scale and billions invested in fab plants across the world. What IBM has done with Theseus is to incorporate a small amount of PCM into a hybrid structure where its ultra-low-latency characteristics can be effectively leveraged. This chart shows the various areas where IBM believes phase change memory could be useful. Note that in many cases, the PCM is being integrated either as a cache solution or as an additional tier of storage between NAND and DRAM, just as NAND is often integrated between DRAM and a conventional hard drive. Project Theseus is an aggregate controller featuring what appears to be 2.8GB of PCM (36 128Mbit cells per card, 5 cards total). IBM calls this its PSS (Prototype Storage Solution). The advantages of PCM are illustrated in the slides above. These graphs show the total latency for various types of requests. Note that the PSS solution (that’s the PCM card) completes the overwhelming majority of its requests in under 500 microseconds. The two MLC solutions top out at 14,000 and 20,000 microseconds compared to 2,000 microseconds for the PSS, while the TLC NAND is an order of magnitude slower, topping out at 120,000 microseconds. In short, these early PCMs, built on 90nm CMOS and at extremely low density (modern NAND flash is now available in 512Gbit sizes compared to 128Mbit for PCM) is a full order of magnitude faster than commercial NAND, with vastly superior write performance and data longevity. There’s just one little problem IBM makes a point of noting that its PSS solution uses 90nm memory produced by Micron. The only problem? Micron gave notice earlier this year that it was cancelling all of its PCM production and pulling out of the industry. While it left open the door to revisiting the memory tech at some point in the future, it indicated that the superior scaling of 3D NAND was a better option (despite the numerous problems identified with that technology in the short term). Where does this leave PCM? The 2013 ITRS report notes that NAND performance isn’t actually expected to increase much from present levels — in fact, it’s going to be difficult to maintain current NAND performance while improving density and holding write endurance constant. Right now, PCM is the most promising next-generation memory technology on the market — but if no one steps forward to manufacture it, it’s going to be a tough sell. Source