29 May 2012—Seeking denser, faster alternatives to the flash memory in smartphones and digital cameras, the world’s largest memory chipmakers are turning to resistive RAM, or RRAM. Unlike silicon-based flash, resistive memory is made with oxides of other metals that can reversibly change their resistances. But silicon isn’t ready to step aside just yet.
This month, a European research team reported a silicon oxide–based resistive memory in the Journal of Applied Physics. It’s the second such device reported: James Tour’s group at Rice University reported one in 2008 and again in the journal Nano Letters in August 2010; the group announced a transparent version of the memory this year at the American Chemical Society meeting in March.
The discoveries came as a surprise to both teams, which made them independently and by accident. Who would have thought that silicon oxide could switch its resistance? After all, its less-silicon-rich cousin, silicon dioxide, has been under chipmakers’ noses for decades—it’s the most common insulator in today’s chips. The new devices still have some weaknesses and face a long road to commercial success, but the implications are big: a dense, scalable, fast, cheap, next-generation memory based on silicon.
“Silicon-based resistive RAM can exist, and that’s really exciting,” says Eric Pop, an electrical and computer engineering professor at the University of Illinois at Urbana-Champaign.
Flash and other major memory types store bits as charge in transistors. Flash is perfect for mobile devices because it’s rugged and nonvolatile—it holds charge when the power is switched off. But the transistors in flash cannot shrink much further: The feature size in today’s memory cells is 20 nanometers wide, making it tough to pack more bits into the same area. Flash is also expensive and slow compared with the dynamic RAM (DRAM) used in computer main memory.
RRAM, in which an amount of resistance represents a bit, is a promising alternative to flash and DRAM, say experts. It relies on memristors, devices that switch their resistance depending on the direction of voltage applied and hold this resistance when the power is off. Memristors could be stacked in three dimensions on the same chip, unlike flash. They are also faster and consume less energy by a factor of hundreds when switching a bit.
Hewlett-Packard demonstrated the first practical memristor in 2008, using titanium dioxide. The company plans to unveil a commercial product in 2013 with Korea-based SK Hynix, the world’s second-largest memory chipmaker after Samsung Electronics, which is planning its own RRAM launch the same year.
Memristors typically contain metal oxide sandwiched between metal wires. An applied voltage makes the normally resistive material more conductive, an effect that can be reversed by flipping the voltage bias.
When Jun Yao, a graduate student in Tour’s lab at Rice, found that silicon oxide does the same trick, the researchers were flabbergasted. Yao sandwiched silicon oxide between polycrystalline silicon electrode layers. He then applied voltage to strip oxygen from the oxide, creating conductive silicon filaments. The filaments could be broken and re-formed by reversing the voltage.
About two years later, Adnan Mehonic, a graduate student in Anthony Kenyon’s lab at University College London who was working on silicon oxide LEDs, saw current fluctuations when he applied a changing voltage across the devices. Careful scrutiny showed that the oscillations were actually a switching between resistances. “We’d thought the device was unstable,” Kenyon says. “I guess we just got lucky.”
When the UCL researchers learned of Tour’s work and realized the potential of their discovery for RRAM, they were excited, particularly since the devices were different.
In the Rice device, conductive filaments are formed on the sides of the sandwich structure. The conductive silicon on the exposed sides can readily mix with oxygen in the air to become resistive again, so the researchers must vacuum-seal the devices. By contrast, the European researchers grow silicon oxide films composed of nanoscale columns of the material packed tightly together. The conductive paths are formed along the columns’ walls, where they are protected from oxidization. The UCL device is the more practical of the pair because it doesn’t require a vacuum, Kenyon says.
The two memristors are each a hundred times as fast as flash, switching in less than 100 nanoseconds. The UCL devices take one thousandth of the energy flash requires to switch. The Rice team has shown that its device can be rewritten 10 000 times, which is comparable to flash. And perhaps most exciting, the conductive filaments in both devices are just 5- to 10-nm wide, suggesting memories at least twice as dense as flash.
But these devices are much further back in the pipeline than the RRAMs planned by Hynix, HP, and Samsung. And they have some big hurdles to overcome, according to the University of Illinois’s Pop. The silicon memristors are big compared with other designs, they need relatively high voltages to switch their resistance, and their reliability is far from proven, he says. “They need to show millions of devices reliably switching resistance hundreds of thousands of times,” Pop says.
Nonetheless, says Tour, “the materials are as simple as can be: just silicon and silicon oxide. I hope that an entire community forms around this methodology, propelling it into commercial applications.”
A correction to this article was made on 30 May 2012.
About the Author