Faster and stronger: next-generation memory tech outperforms flash

Via ars technica

-----

Flash memory is the dominant nonvolatile (retaining information when unpowered) memory thanks to its appearance in solid-state drives (SSDs) and USB flash drives. Despite its popularity, it has issues when feature sizes are scaled down to 30nm and below. In addition, flash has a finite number of write-erase cycles and slow write speeds (on the order of ms). Because of these shortcomings, researchers have been searching for a successor even as consumers snap up flash-based SSDs.

There are currently a variety of alternative technologies competing to replace silicon-based flash memory, such as phase-change RAM (PRAM), ferroelectric RAM (FERAM), magnetoresistive RAM (MRAM), and resistance-change RAM (RRAM). So far, though, these approaches fail to scale down to current process technologies well—either the switching mechanism or switching current perform poorly at the nanoscale. All of them, at least in their current state of development, also lack some commercially-important properties such as write-cycle endurance, long-term data retention, and fast switching speed. Fixing these issues will be a basic requirement for next-gen non-volatile memory. 

Or, as an alternative, we might end up replacing this tech entirely. Researchers from Samsung and Sejong University in Korea have published a paper in Nature Materials that describes tanatalum oxide-based (TaO_x) resistance-RAM (RRAM), which shows large improvements over current technology in nearly every respect.

RRAM devices work by applying a large enough voltage to switch material that normally acts as an insulator (high-resistance state) into a low-resistance state. In this case, the device is a sandwich structure with a TaO_2-x base layer and a thinner Ta₂O_5-x insulating layer, surrounded by platinum (Pt) electrodes. This configuration, known as metal-insulator-base-metal (MIMB), starts as an insulator, but it can be switched to a low resistance, metal-metal (filament)-base-metal (MMBM) state.

The nature of the switching process is not well understood in this case, but the authors describe it as relying on the creation of conducting filaments that extend through the Ta₂O_5-x layer. These paths are created by applying sufficiently large voltages, which drive the movement of oxygen ions through a redox (reduction-oxidaton) process.

When in the MIMB state, the interface between the Pt electrode and the Ta₂O_5-x forms a metal-semiconductor junction known as a Schottky barrier, while the MMBM state forms an ohmic contact. The main difference between these two is that the current-voltage profile is linear and symmetric for ohmic but nonlinear and asymmetric for Schottky. The presence of Schottky barriers is a benefit, as it prevents stray current leakage through an array of multiple devices (important for high-density storage).

The results presented by the authors appear to blow other memory technologies out of the water, in pretty much every way we care about. The devices presented here are 30nm thick, and the switching current is 50 ?A—an order of magnitude smaller than that of PRAM. They also demonstrated an endurance of greater than 10¹² switching cycles (higher than the previous best of 10¹⁰ and six orders of magnitude higher than that of flash memory at 10⁴-10⁶). The device has a switching time of 10ns, and a data retention time that's estimated to be 10 years operating at 85°C. This type of RRAM also appears to work without problems in a vacuum, unlike previously-demonstrated devices.

This may all seem too good to be true—it should be emphasized that this was only a laboratory-scale demonstration, with 64 devices in an array (therefore capable of storing only 64 bits). There will still be a few years of development needed before we see gigabyte-size drives based on this RRAM memory.

Like all semiconductor device fabrication, advances will be needed to improve nanoscale lithography techniques for large-scale manufacturing and, in this particular case, a better understanding of the basic switching mechanism is also needed. However, based on the results shown here, this new memory technology shows promise for use as a universal memory storage: the same type could be used for storage and working memory.