Resistance is not futile
Can metal oxides solve a 40 year problem for memory makers?
Chipmakers have tried for years to find the perfect memory: something that is fast, cheap and remembers what you put into it when you power down. Despite 40 years of trying, nothing exists on the market that gives you three out of three.
There are very few electronic systems out there that can get away with just one form of memory inside for storing data. Power meters and radio-frequency tags make use of ferroelectric memory to keep data for long periods of time without having to power it. Even in those systems, the code will sit in flash memory or even cheaper mask-programmed memory because the ferroelectric memory cells are comparatively expensive.
Twenty years ago, ferroelectric memory looked set to be the memory of the future. But problems with manufacture made it tough to scale up densities. It started off a couple of generations behind mainstream DRAM and flash in terms of capacity. While DRAM and flash surged into the gigabit era, ferroelectric memory makers struggled to get above a megabit.
In recent years, ovonic or phase-change memory made a comeback. This was a universal memory from the dawn of the integrated circuit. Gordon Moore worked on it in the late 1960s but its shortcomings outweighed any advantages that might come from continued development. Intel took another look at phase-change several years ago and got to the stage of sampling test devices to cellphone makers before the research was spun out to the Numonyx joint venture set up with STMicroelectronics, which had its own programme.
NXP Semiconductors has also continued to work on phase-change memory, and is able to call on research from former parent Philips Electronics, which developed a number of the materials used.
Now attention is settling on yet another candidate for the perfect memory: the resistive RAM. Researchers have found that metal oxides can be made to store data and hang onto that data after the power supply has been disconnected. They are hoping this one works as there is no guarantee that the mainstays of the memory business can keep growing.
MJ Lee, an engineer with Samsung Semiconductor, says the company sees resistive RAM as an alternative to flash: "The flash market grows remarkably each year but the floating-gate technology is reaching its fundamental limit of density. The NAND flash will reach its scaling limit around the 30nm or 20nm process node. Resistive RAM is a strong candidate for the next-generation memory."
The big advantage of the
resistive RAM is its relative simplicity. The successful memories call for unusually shaped structures to be built on the surface of the silicon. DRAM needs a capacitor which, to keep the stored charge high, needs more exotic shapes. Flash demands extra layers be built into a tiny transistor.
The resistive RAM just puts a layer of oxide at the intersection between two metal strips. If the oxide layer passes a current, that is read as a '1'. If it acts as a resistor, then the read circuitry sees that as a '0'. Lee says the resistance ratio in Samsung's experimental memory is more than two orders of magnitude.
You switch a resistive RAM in a similar way to phase-change or magnetoresistive memory, another one-time perfect-memory candidate: by passing a high current through it. One advantage that resistive memory has is that the switching time is much faster than is possible with flash. It takes less than 10ns to set a bit; about 10ns to reset it.
Because the RAM relies on the ability to measure changes in resistance, Lee says multilevel memories, similar to the technique used in NAND flash, are possible. A further possibility is building multilayer memories - it is just a matter of adding pairs of metal layers and circuitry to address each layer.
Samsung developed a two-layer stack last year. The biggest problem, Lee says, was making sure that the nickel oxide layer used in the device survived the high temperatures need to lay down the upper metal matrix. "The nickel oxide can be grown under room-temperature conditions," he claims. "Our results show that, during operation, there is no interference between the layers."
There is, however, a catch. In fact, there are a couple. Ugo Russo of Milan Polytechnic says: "The big problem is the relatively high current needed to perform the reset. It suggests that Joule heating plays a role in the reset operation."
The problem is similar to that of phase-change memories, which also rely on the heat generated by electrical current passing through the material to switch. This kind of power is bad news for battery-powered systems.
"We investigated if we can, in some way, reduce the current. As we reduce the diameter of the oxide filament, the current reduces. That reduces both the electrical and thermal resistance of the filament," Russo claims. "The question is how to control the size of the filament."
Katsuhisa Aratani of Sony says one approach that reduces switching current is to apply a negative voltage to the bottom electrode and use pulses of current. However, he says a further problem is storage time: "The memory is better for fast switching compared with other non-volatile memories but it is not good for retention. Something needs to be added to improve the retention time."
Because the reset process seems to rely partly on heat, it is possible for the memory to reset itself slowly at room temperature. Work with materials allowed Sony to push up the retention time to around ten years at room temperature based on tests at 130°C.
"It is promising for a next-generation non-volatile memory," says Aratani, who reckons it will scale down well to the 10nm generation at least.
Before resistive RAM can make it to market, it will have to go through many more tests and manufacturability work. These are the areas that have tripped up earlier candidates for the perfect memory. So, you will probably have to wait a while longer before you can kiss goodbye to DRAM and flash.