(Invited) Is Ion Migration Needed for Resistance Memory and Memristor?

Tuesday, 3 October 2017: 16:30
Camellia 4 (Gaylord National Resort and Convention Center)
I. W. Chen (U Penn, Dept Materials Science & Engineering)
This presentation will focus on the electronic behavior of resistance switching memory, and contrast it with the expectation of ionic and Joule heating mechanisms that have received much attention in this field. Several prototypical memories, ranging from the filamentary type to the nanometallic type, will be examined. Filamentary memory is believed to undergo resistance switching along a nanoscale filament between the electrodes, and there is extreme concentration of ion and heat flow along the filament. It usually requires dielectric breakdown (called “forming”) prior to operation, or else the device remains resistive and cannot be switched. Nanometallic memory is believed to undergo resistance switching only when the film is thinner than a certain thickness, and it switches uniformly by way of voltage-trapping/detrapping of electrons, which alter the localization length of electron wave function. Although their switching and conduction mechanisms have been thought to be entirely different, one relying on ions, the other electrons, there are surprising similarities between the two, summarized below.

*Their high resistance states are insulating, and their low resistance states are metallic, having opposite signs of temperature sensitivity of resistance, as evidenced by resistivity data down to 18 mK.

*Their high resistance states are unstable under a mechanical pressure, which triggers them to switch to the low resistance states in less than a picosecond without any voltage assistance.

*Their low resistance states are stable under a mechanical pressure.

*Their on/off switching is triggered when their switching elements experience a critical voltage, which because of load sharing can be substantially less than the device voltage.

Since as-fabricated nanometallic memory is already in the low resistance state, it is also stable under a mechanical pressure. However, a virgin, highly resistive filamentary memory can be alternatively “formed” to the low resistance state by a pressure alone, without going through dielectric breakdown. This, along with pressure-triggered switching from the high resistance state to the low resistance state, provides unequivocal evidence that forming and switching can proceed without any ion migration, even in a filamentary memory.

In addition to the above, voltage-controlled switching is difficult to rationalize by the existing switching models for filamentary memory. These models all invoke a critical amount of electric current and joule heating to break and reconnect filaments, which are not likely to be voltage-controlled processes.

Lastly, filamentary memory has been used to emulate synapses. But all synaptic ion channels (Na or Ca) are voltage-gated. Therefore, the finding that both nanometallic and filamentary memories are voltage-controlled is fundamentally relevant to using memristors for neuromorphic computing.