Jean-Pierre Braun, the CEO of Crocus technologies, is visiting Israel these days, and I had the good chance of meeting him. Crocus' technology is based on work done by the French Spintec research center. In fact they get an exclusive access to Spintec MRAM related research and patents for the next 15 years.
Crocus' basic technology is based on Field-Induced MRAM (Toggle MRAM), which is in some ways similar to Everspin's technology (or 1st generation MRAM). Historically, Field-Induced MRAM is very hard to scale, and has stability and retention problems. Crocus aims to solve all this using a thermally activated magnetic latch. They call their technology Thermally Assisted Switching or TAS. Basically this means that the latch helps the cell retain the memory value very well. It can also scale quite well. This also (theoretically) will allow them to pack more than 1 bit per cell... some day.
Jean-Pierre says that many companies are doing 'Thermal MRAM' - which only means heating the material for easier handling. But Crocus technology is very different - because of the use of this magnetic latch, not just heat by itself.
Crocus already has a 'small' (few KBits) working sample of their memory, and towards the end of 2009 they will hopefully get a real product out. I can't say much yet, but this product will have a higher-density than Everspin's products (currently 4Mbit), smaller size and will also be considerably cheaper. This is great news, and hopefully they'll be able to pull it off. Even though the largest MRAM market is for embedded memory, the first Crocus product will be stand-alone.
Obviously these are hard times - especially for start ups and companies that are still losing money. Crocus has already raised around US$30M, but they aim to break-even in 2010 or 2011. They will still require another round of investment, but hopefully a small one.
Crocus are also working on STT-RAM. Jean-Pierre thinks that the best way forward is Perpendicular STT-RAM, which is currently lead by Toshiba. Crocus' MRAM road map looks like this -
- 130-90ns : Field induced TAS MRAM
- 65-45ns : STT-MRAM
- <45ns : STT-MRAM with TAS for stability reasons
But volume production of STT-RAM is still way off, probably "at least 4-5 years" away according to Jean-Pierre.
Will MRAM ever be the 'Universal Memory'?
In the happy days of 2004, MRAM was hyped as the universal memory - being able to replace your SRAM, DRAM, FLASH and H/D - all in one chip. This is great - but not realistic - at least Jean-Pierre does not believe it. It will require some serious technological breakthrough to compete in particular with NAND Flash, and he just can't see it happen anytime soon. Better to focus on 'niche' memory markets that will take full advantage of the specifics of MRAM - infinite endurance and fast write & read cycles, reliability, small die size and low memory requirements. I'm also quite tired of the old MRAM buzz and hype, and hopefully Crocus and other companies are more sober today and this might actually lead to commercial products. Good luck!