Heat flash! Current through the gate layer creates heat that "heals" the oxide layer. (via Macronix and IEEE)
Spintronic memory may take over in the future, but until then, we need to extend the life of what we do have available. Flash memory is some of the most reliable, non-volatile memory, but it tends to fail after about 10,000 write/erase cycles. However, engineers from the Taiwanese company, Macronix, have found a way to anneal the parts damaged in NAND flash cells and bring them back to working order after the 10,000 cycle limit.
A flash memory cell works by storing one bit of electrons on a “floating gate,” which is embedded between two insulators. After too many cycles, this insulation becomes degraded, and when it does, the cell stops functioning properly. Computer engineers know this happens, and so they make sure to strategically store data so that cells are not used more repeatedly than they have to.
The Macronix team, lead by engineer and deputy director, Hans-Ting Lue, was inspired by Phase change RAM (PCRAM) which uses chalcogenide glass, that when heated, changes from an insulator to a conductor. They saw that heating these chips could “heal” them and they decided to adapt this “healing” to flash memory.
Macronix engineers found that annealing BE-SONOS (silicon-oxide-nitride-oxide-silicon) NAND Flash chips worked with just milliseconds of heating. In these chips, the nitride floating gate is insulated by surrounding oxides, which are what degrades after being submitted to strong fluctuations in the surrounding electric field.
Engineers already knew that heating this oxide insulation could anneal it or “heal” it but it was believed this had to be done to the entire chip at temps around 250C for hours. Obviously, this method was impractical. The team decided to redesign a BE-SONOS NAND Flash chip so it could simply heat this section of oxides around the floating gate.
To achieve this feat, they adapted a chip with new gate electrodes i.e. “changed the wordline from a single-ended to a double-ended structure”, that could carry the few millisecond-long jolt of current necessary to heat the cell’s insulation to more than 800C. They also equipped the chip with additional diodes, all of which resulted in a much bigger chip; however, the Macronix team was able to redesign the chip’s architecture to fit it all in a similar space.
This heating could be planned so that it only heals tiny sections of the chip at a time and seldom happens. Devices would need to be turned off, but the heating could take place when the main device is charging so these self-healing chips will not wear out battery life.
The heating was found to be delivered highly effectively and exclusively to the point of interest, and the rest of the chip was not affected. Unexpectedly, engineers also observed that heating while erasing resulted in faster erasing speeds because, at this extreme temperature, electrons were de-trapped from the nitride floating gate so Fowler-Nordheim tunneling, occurring during draining of the memory cell, happened faster.
Macronix researchers believe these self-healing chips could last up to 100 million write/erase cycles, though this limit has not been reached so it could be larger. Lue said a test of one billion cycles would take several months and has not yet been performed.
This self-healing flash, although not the same as RAM, could see a redesign in its architecture and could eventually replace dynamic RAM. This type of Flash memory chip could also be particularly useful in server farms, where hard drive failure is common. However, I would imagine that solid state drive prices would still keep rotary/mechanical drive prevalent for some time.
Lue was reluctant to discuss when we could see a self-healing flash chip on the market. Macronix already manufactures NOR Flash, NAND Flash and ROM products so the lack of urgency is understood. Macronix will have a presentation at IEEE’s International Electron Devices Meeting that will be held from December 10th to the 12th in San Francisco. News of their new memory chip was published in IEEE Spectrum.
When commenting on Macronix’s method for restoring Flash memory, Lue said, “It took a leap of imagination to jump into a completely different regime… very high temperature and in a very short time. Afterward, we realized that there was no new physics principle invented here and we could have done this 10 years ago.”
Cabe