Engineering For Next-Gen Memory Performance

Progress is being made on RRAM, but it’s still not ready for prime time.


When only a few electrons mean the difference between the ON and the OFF state, it’s difficult to manufacture memory elements with consistent, reliable performance. This is the situation conventional capacitance-based memories face as critical dimensions drop to just a few nanometers.

As a result, device designers are considering a wide range of alternative memory elements. One of the simplest of these is Resistive RAM (RRAM), also known as crosspoint memory for one proposed implementation. In RRAMs, a metal-insulator-metal structure has high resistance in the OFF state, but low resistance in the ON state. There are a number of proposed material systems for these devices; one of the most studied places a layer of HfO2 between two TiN electrodes. In a crosspoint architecture, the top and bottom wire layers are at right angles, creating a grid. Memory elements lie at the corners of the grid.

As researchers at SEMATECH reported in 2011, oxygen vacancies in the HfO2 layer segregate to the grain boundaries. When a sufficiently high current is applied, the grain boundaries form a percolation path, allowing electrons to propagate through the material by trap-induced tunneling. Hf-O bonds along this path break and Hf-Hf bonds form, ultimately creating a metallic conducting filament between the top and bottom electrodes. This is the ON state. Because it does not depend on stored charge, it is extremely stable.

To reset the memory to the OFF state, the system applies a current, increasing the temperature of the hafnium filament and causing it to re-oxidize. Ultimately, the complete conductive path is lost and the dielectric again behaves as an insulator.

RRAM implementations have struggled with high variability and inconsistent performance. The distribution of grain boundaries in the material is difficult to control at the nanometer scale. Moreover, the dielectric breakdown mechanism is inherently variable, depending on stochastic movements of electrons between defects.

At this year’s Materials Research Society Spring Meeting in San Francisco, researchers offered several proposals for reducing the variability of RRAM devices. One of the more interesting ideas came from Joshua Yang at Hewlett-Packard Laboratories. He observed that transport of hafnium and oxygen ions within the HfO2 material is complex, involving drift, electromigration, diffusion, and other transport mechanisms. Instead, he suggested, the insulating layer can be seen as composed of a hafnium-rich conducting phase and an oxygen-rich insulating phase. By creating a composite, with hafnium-rich nanocluster “seeds” embedded in the HfO2 matrix, applied current can drive a phase change from one to the other.

Alternatively, SEMATECH’s Gennadi Bersuker observed, an oxygen-deficient HfO2 layer offers a pre-existing supply of oxygen vacancies, and therefore a pre-existing source of dangling hafnium bonds. If the oxygen vacancy concentration is known, the device switching characteristics are easier to predict.

RRAMs have clear advantages in their simplicity, scalability, and stability. Consistent performance is essential for commercial devices, though. These results show that progress is being made, but also that much work still remains.