ARM Holdings plc (Cambridge, England) is turning its attention to non-volatile memory technology and that’s good thing. Recent announcements have covered both magnetic RAM and correlated electron RAM and show ARM stepping up and helping to drive research that could move the electronics industry forward.
ARM is a well known as a supplier of intellectual property for processors. It is also well known as not being a manufacturer of its own chips and indeed not even a designer of complete chips. It is both fabless and chipless – and licenses processor cores to fabless and IDM chip companies in return for initial fees and royalties.
As such it would be easy for ARM to insist it is always some other company’s responsibility to develop a manufacturing process, research a better transistor and develop a better memory. And in general that has been ARM’s position through its more than 20 years of existence.
But as ARM has achieved greater global success the level of circuit integration has increased. Back in 1990 low-power logic through elegant minimal design was ARM’s key benefit but significant amounts of memory were always an off-chip consideration. Subsequently the sizes of on–chip cache have grown and in contemporary application processors and SoCs multiple processor cores must access multiple on-chip memory spaces. Burning power in those memories is starting to become highly detrimental to efficient performance. Similarly moving memory contents on- and off-chip is power hungry. This all points to on-chip non-volatile memory available in a logic manufacturing process – or logic in a non-volatile process – as something very much to be desired.
Fundamental device capability has always linked to manufacturing process and design but in the past the biggest differentiator was design. Now, just as almost everyone and their dog is getting out of manufacturing and process development (see IBM) the balance could be tipping back towards knowing the ins and outs of quantum mechanically influenced transistor and memory structures and being able to manufacture them.
I don’t expect ARM to change its policy towards manufacturing but the company has clearly grown up and now feels it is part of its responsibility to help develop all the essential capabilities, both in its own best interest and in the interest of its electronics ecosystem.
The first and nearest term announcement was that ARM, which usually licenses technology out, has licensed in MRAM from startup Crocus Technology SA (Grenoble France) to aid the development of security and microcontroller chips.
Crocus is providing ARM with access to its Magnetic Logic Unit technology including sub-90nm MRAM blocks that can replace traditional Flash based non-volatile memory, and match-in-place technology that Crocus claims enhances the security of keys and other secret data.
Vincent Korstanje, vice president of marketing, systems and software at ARM, said at the time: “The potential impact of emerging non-volatile memory technologies is substantial. ARM is keen to investigate and understand how this may ultimately affect our broad ecosystem of silicon partners, OEMs and other stakeholders.”
Further from market but also of potentially great significance is the news that ARM is evaluating the correlated electron RAM (CeRAM) technology being developed by Professor Carlos Paz de Araujo and colleagues at the University of Colorado and Symetrix Corp. (Colorado Springs, Colo.).
The claim is that the CeRAM is a non-filamentary, non-electroforming resistive RAM that allows robust, high-temperature storage showing 400 degrees C prolonged storage without degradation, read voltages at 0.1-0.2 V, and endurance cycling above 10^12 cycles. Early indications are that CeRAM could fit with CMOS process flows and monolithic 3D stacking. The non-filamentary claim is key as it suggests the memory could scale below 10nm dimensions.
According to Symetrix ARM is evaluating the technology as part of its strategy in the embedded nonvolatile memory offerings and Symetrix has agreed to provide results to foundries engaged by ARM. Professor Araujo has told me that ARM will use device models for array design and cell development. He added that CeRAM could be in commercial production in embedded applications within two years, such as non-volatile memory for a microcontroller on a modest node such as 130nm.
Displacing flash in stand-alone stacked memory would probably take longer, but it is good that ARM is taking a lead in researching these things.
It is well known that getting the direction of technology or market development is relatively easy. It is getting the timing right that makes the difference between success and failure. In my questioning of ARM executives over recent years I have been repeatedly asking about non-volatile memory. It now feels ARM is playing all the right notes and probably in the right order – and with a little accelerando.
Related links and articles: