Materials For Future Electronics

Examining the research underway in electronics materials provides a keyhole view into what may be possible in future electronics design. Although some of this research will not end up in commercial products, it does provide an indication of the kinds of problems that are being addressed, how they are being approached, and where the research dollars are being spent.

Flexible electronics are attracting a steady stream of research because of potential applications in wearable electronics and devices. While researchers at the IEEE Electron Device Meeting in December focused on hybrid structures, with silicon electronics attached to flexible substrates, papers presented at the recent MRS Spring Meeting also considered the integration of flexible organic semiconductors.

The nature of the proposed applications imposes very challenging tradeoffs between flexibility and performance, though. Zhenan Bao, professor of chemical engineering and material science and engineering at Stanford University, explained that devices intended to move with the torso must be able to tolerate as much as 50% strain. Inside-the-body applications, such as monitors attached to the heart, see 20% to 35% strain. Even the brain expands and contracts, imposing up to 10% strain on sensors attached to it.

While organic semiconductors are more flexible than silicon, the materials with the highest mobility generally have a rigid, crystalline molecular structure. Within the material’s tolerance, increasing strain in general still will decrease mobility. Bao’s group used selective “breaking” of the conjugated carbon backbone structure to reduce the modulus of indacenodithiophene (IDT)-based semiconducting polymers while maintaining carrier mobility.

Fig. 1: Flexible nanonet circuits. Source: Purdue University

Another longstanding challenge for flexible electronics concerns the interface between organic sensors and semiconductors, and conventional metallic electrodes. Changhee Lee, professor in the Department of Electrical and Computing Engineering at Seoul National University, and Chan-mo Kang of the Electronics and Telecommunications Research Institute in Seoul, used self-assembled monolayers to tune the surface behavior of gold electrodes, lowering the injection barrier and improving through-plane mobility at the interface between gold and pentacene.

Materials enable advanced memories
Elsewhere in the electronic materials world, research continues on two novel memory concepts, phase change memory and ReRAM. In both cases, data is stored in the form of a reversible change in resistance, not as a capacitive charge. As a result, these memories can support very simple, compact architectures.

In phase change memory, an electrical pulse heats the material, causing a transition from an insulating amorphous phase to a conducting crystalline phase. Because the transition is thermally-driven, care must be taken to avoid heat transfer between adjacent cells. Moreover, the performance of memories based on these cells is very sensitive to the switching kinetics, which are not well understood.

In the standard view, Aaron Lindberg, associate professor of materials science and engineering at Stanford University, explained that the amorphous phase begins to conduct electricity at some threshold electric field, allowing resistive heating, which in turn initiates the phase transition. As energy pulses — supplied either electrically or optically — get shorter, the power needed to switch the device increases, subject to some asymptotic limit below which no transition occurs.

However, experiments with picosecond and femtosecond pulses have shown that this model is too simple. Lindberg’s group found that electric field pulses raised the temperature of AgInSbTe (AIST) by only about 0.6° K, while its crystallization temperature is close to 440° K. Anbarasu Manivannan, associate professor of electrical engineering at the Indian Institute of Technology, reported that the electrical transition from the “off” to “on” state in AIST occurred in only 250 ps, but estimated the crystallization time at 700 ps. In both cases the electrical transition appears to proceed ahead of the phase transition, so some additional mechanism appears to be involved.

ReRAM devices do not depend on a phase change, and appear to have simpler kinetics. There are two main ReRAM types, both of which depend on oxidation-reduction (REDOX) reactions in metal-insulator-metal structures. In valence change memristor (VCM) devices, the movement of oxygen vacancies in the metal-oxide insulating layer creates a conductive filament, switching the resistor on. In electrochemical memristor (ECM devices), oxidation at one of the metallic electrodes leads to filament formation.

ReRAM devices were first considered as a possible replacement for NAND flash, which at the time appeared to be reaching its scaling limits. ReRAM potentially offered a high device density with a low read voltage. However, Dirk Wouters, a researcher at RWTH Aachen University, explained that the set voltage for these devices lies at some level above the read voltage. In low power operation, the set voltage level may be higher than the maximum voltage supplied by the circuit, eliminating the window in which programming can occur.  Meanwhile, the successful deployment of flash structures has postponed the need for an alternative non-volatile memory technology.

Beyond von Neumann computing
The unique behavior of ReRAMs also has attracted interest from a completely different direction—neuromorphic computing. In conventional von Neumann computing architectures, there is a clear distinction between the functionality of the core computing elements and the memory subsystems. In operations on large data sets — such as calculating the weights for nodes in a large neural network — transferring data between the memory and computing elements often introduces a significant performance bottleneck.

Neuromorphic architectures, by analogy to synapses and neurons in the brain, seek to perform computations directly with the memory elements. ReRAM facilitates such architectures because, while conventional memories are either ON or OFF, an ReRAM can have multiple resistance levels, corresponding to the growth of the conducting filament. In theory, just as connections in the brain are reinforced by repeated stimuli, data stored in ReRAM memories can become “stronger” with repeated inputs, or can decay if a stimulus is not repeated.

As Wouters pointed out, though, very few memristor arrays have transitioned from conference presentations to real devices. It’s difficult to evaluate a design without a physical prototype. Dmitri Strukov, associate professor in UC Santa Barbara’s department of electrical and computer engineering, noted that systems based on analog memories have so far failed to match the performance of more conventional designs. The first obstacle to real implementations is very basic indeed—the need for ReRAM devices that offer a gradual and consistent transition from the “off” to the “on” state.

In most ReRAMs, the transition between the two states is abrupt and binary, apparently due to the completion of a conducting path through the device. Researchers, led by Hewlett Packard Enterprise senior fellow R. Stan Williams, used the scanning transmission x-ray microscopy facility at Lawrence Berkeley National Lab to actually observe the formation of a strong conduction channels in hafnium oxide as oxygen vacancies moved under the influence of an applied field. In order to obtain a more gradual transition, Huaqiang Wu, deputy director of the Institute of Microelectronics at Tsinghua University, suggested that one might seek to form multiple weak paths through the insulator by dispersing oxygen vacancy sites more widely. For example, increasing the temperature of the insulating layer might disperse vacancies more uniformly, or aluminum doping might localize vacancy formation near the dopant sites.

Fig. 2: Hafnium Oxide. Source: Stanford Materials

All of the devices discussed here are highly experimental at this point, with yield, switching characteristics, and the underlying physics all posing significant questions. It will be several years before we know whether ReRAMs can be used to build facial recognition systems that match the human brain, or flexible semiconductors can enable clothing that monitors blood chemistry while an athlete moves. As the integrated circuit industry approaches the potential end of silicon, though, it’s clear that the universe of design possibilities is still expanding.

Related Stories
Progress In Flexible Electronics
Hybrid approach pairs very thin silicon with printed interconnects and sensors.
Many Paths To Hafnium Oxide
What makes a good precursor in atomic layer deposition isn’t clear.
What’s Next For Transistors
New FETs, qubits, neuromorphic approaches, and advanced packaging.
New Memories And Architectures Ahead
So far there is not widespread adoption, but many see change as inevitable.