I was also going to add that we had not heard much from Rambus about their ReRAM portfolio following their purchase of Unity Semiconductor and their CMOx technology. I heard rumours the focus was turning to embedded rather than stand-alone memory and was contemplating some sort of ‘predictive wisdom’ when I was beaten to it! Rambus have signed an agreement with Tezzaron who will develop ReRAM for some niche but high value products aimed at, for example, military and aerospace. Again the focus is performance, rather than cost especially for minimizing power and, interestingly, radiation hardness, so maybe that has been the focus of Rambus’ ReRAM development since the Unity purchase. Bob Patti, Tezzaron’s CTO has been quoted as saying that “the company (Tezzaron, I assume) focuses on creating devices with self-repair features for devices that can’t be easily accessed for maintenance” mentioning the Mars Rovers. I wonder if they have looked at the issue of ‘repairing’ crosspoint cell that gets stuck in a low resistance state as that can knock out a lot of neighbouring cells… First products are projected for 2016 so I’m safe for a 2015 prediction!
Sony/Micron are still (publicly, at least) sounding very bullish about their CBRAM based approach and have mentioned cache like products in the near future with major announcements such as at ISSCC 2014 and FMS. However, there is no follow-on at ISSCC this year and it is clear any such products (in the near future) will be more expensive than those based on existing technology so the performance boost will have to be compelling. There has been interesting work from various academic groups, notably Professor Ken Takeuchi’s group at Chuo University on this topic but of course, a hardware demonstration is yet to appear. And all this is predicated on the assumption that Sony/Micron can solve or have solved the problems of volume manufacturing.
The news from HP last year switched fairly dramatically to ‘The Machine’ and away from updating the roadmap for the availability of their resistive memory. HP has made a bold prediction that an early version of the operation system (‘The Operating System’?) for ‘The Machine’ will be available this year with hardware and memory to come over the next few years. HP have definitely raised the stakes this time with these announcements being made by top Executives along with a redirection of the majority of HP Labs in support of this project. To be perfectly honest it is easy to be somewhat cynical of this type of bold statement given the repeated delays in previously announced schedules. If nothing else it shows a gap between the smart people in HP R&D and those, equally smart I’m sure, in the sharp end of product development. I’m sure HP’s competitors are working on key elements of ‘The Machine’ notably photonic data transfer and ways of improving memory bandwidth (see above). And introducing these technologies into an existing server framework is going to be challenging enough without all the inevitable headaches of designing and producing a brand new operating system that will take full advantage of new technology yet still provide legacy compatibility.
Neuromorphic computing does seem to be progressing, albeit not quite as I ‘foregazed’ a year ago. IBM have designed, fabricated and analysed a 165k synapse chip in a 180nm technology using PCM (phase change memory) devices in a 1T1R configuration, reporting the results at IEDM and no doubt elsewhere. The resistances of the PCM elements are tuned to adjust the weights between the nodes of an artificial neural network (ANN). Indeed, I remember (reasonably) fondly the participating in the big resurgence of ANNs in the late eighties and early nineties. Note this is rather different from more recent approaches to neuromorphic computing such as Spike Timing Dependent Plasticity (STDP, about which I admit to knowing very little) and implementations exploiting the integrating capability of a resistive memory element. Nonetheless, I was expecting a bit more in terms of breakthroughs in this area last year. So I will boldly defer my 2014 prediction to 2015 (and here I am being somewhat critical of HP for doing more or less the same thing…, sorry guys and gals at HP).
(That’s enough predictions, Ed)
Christie Marrian, Director, ReRAM Forum