Toshiba's New STT-MRAM Beats SRAM Power Drain by 90%
Toshiba has developed a new element for MRAM, lowering its overall power consumption.
Toshiba said on Monday that it has developed a prototype memory element for a spin transfer torque magnetoresistive random access memory (STT-MRAM) that (currently) achieves the world's lowest power consumption.
According to the company, the new MRAM element has the potential to surpass the power consumption efficiency of SDRAM as cache memory in practical operations. Toshiba confirmed the performance by using a highly accurate processor simulator, recording a two-thirds reduction in power consumption by a standard mobile chip set carrying out standard operating functions.
"Going forward Toshiba expects to bring the new memory element to STT-MRAM cache memory for mobile processors integrated into smartphones and tablet PCs, and will promote accelerated research and development toward that end," the company said.
Mobile devices like smartphones and tablets depend on high-speed memory to supply the main processor with instructions and frequently requested data. Typically SRAM is used as a cache-memory solution, but improving its performance to match rapidly advancing products has led to increased current leakage in standby mode and during operation, degrading power performance.
Thus, enter MRAM, which is based on magnetic materials. MRAM was introduced as an alternative to SDRAM because it is non-volatile, cutting leak current during standby status. However until now, the drawback to using MRAM was that power consumption exceeded that of SDRAM, making it less ideal for practical applications. The new element developed by Toshiba changes all of that.
"Toshiba's new memory element advances the company's pioneering work in STT-MRAM and overcomes the longstanding operating trade-off by securing improved speed while reducing power consumption by 90 percent," the company said. "The improved structure is based on perpendicular magnetization3 and takes element miniaturization to below 30-nm. Introduction of this newly designed "normally-off" memory circuit with no passes for current to leak into cuts leak current to zero in both operation and standby without any specific power supply management."
Toshiba said it will present three papers on the new STT-MRAM and its technologies on December 11 and 12 at IEDM, the International Electron Device Meeting held by IEEE in San Francisco.
We use SDRAM in memory because it is far denser, and thus cheaper, not because it is more power efficient. For example, the Bulldozer and Piledriver dies are made of mostly SRAM cache, yet they're like two orders of magnitude less dense and it's not even the least dense form of SRAM that the dies mostly contain in theL2 and L3 caches that together take up about three fourths or four fifths of each die.
Quite the opposite - SRAM needs little energy compared to DRAM - this is what the battery on your motherboard is for (keeping BIOS config data in SRAM) and it will last years doing so.
SRAM cell requires 6 transistors while DRAM needs only 1 transistor and 1 capacitor so it's much cheaper for the same capacity. See what blazorthon wrote.
Just give a couple of hundred million to a university to found a research program. What, do you think companies are not smart enough to be working on it if there was a profit to be made? Or do you think they are holding it back out of spite?
There could easily be a small magnetic shield integrated into the exterior of the chips. It's not as bad as hard drives about magnetic fields IIRC. Besides, most people aren't in situations where very strong magnetic fields are a problem. How many people have laptops with hard drives, yet never complain about their hard drives getting wiped by magnetic fields around them?
I've been reading about a few MRAM technologies for years... They haven't gotten very far. Most usage that I'm aware of is small capacity chips that are used as replacements for SRAM chips that need non-volatility. It might get somewhere thanks to this development, but I'm skeptical. Most technologies like it have yet to get anywhere despite decades of R&D. I hope to be proved wrong because it's quite an interesting and potentially very practical technology for a huge variety of usage.