What makes a metal conducting and an insulator insulating? This question is easily answered for metals that form band gaps with sharp edges, but the answer is unclear for amorphous materials that have many electron states inside the gap. In 1958, Anderson proved that when the disorder exceeds a certain critical value the wave function exponentially decays. So an insulator is a disordered electron system in which an electron cannot diffuse without bound; in the disordered landscape it can only reach a finite distance, called localization length. This concept inherently entails a size effect, which was first demonstrated in 2005 by Chen’s group at the University of Pennsylvania: disperse metal atoms embedded in a disordered insulator can provide electrons that communicate over a nanoscale distance. We now know that there is a large family of metal-doped amorphous oxide and nitride thin films that behave like conductors when the film thickness is less than the localization length. Moreover, their localization length is tunable by electrical voltage, so when these films are electrically switched between nano metals and insulators, 1 and 0 states are encoded. These nanometallic resistance memories have provided a new material platform to make simple two-point devices known as memristors.
When I joined the group in 2012, I was asked by Prof. Chen: “What do you think should be the next nano metal to study, the ultimate material for memristive devices?” Silicon was the word that came out of my mouth. Trained as an electrical engineering undergraduate, I knew it best, it is the keystone of electronic industry, and to me it seems such a simple, basic material. “Sounds very interesting. Let’s make it work.” So it became my PhD project. Unbeknownst to me at the time was that it is not “basic” at all to turn silicon into a memristor, and in fact several former group members knowing a lot more materials science than I did not take on the project because they didn’t believe it was possible—or else it would have been done a long time ago. Being naive at the beginning, I first followed the same approach of doping metals into amorphous silicon hoping to see hysteresis in the current-voltage responses, but it only made the film more conducting and there was no sign of a memristor. In fact, I found silicon without doping is already nanometallic when the film is thin enough. From those failures, we realized that unlike amorphous oxides and nitrides in which metal is needed to reduce the localization length, silicon having a tighter band gap and already a nanometal needs tuning in the opposite direction, i.e., to increase the localization length. In the summer of 2013, the breakthrough finally came in sample #179, an amorphous silicon lightly doped with oxygen was hysteretic. I finally turned the basic silicon into a memristor.
After filing the patent for the discovery, I spent the next few years probing electrons in pure silicon and O/N doped silicon to prove that they are indeed nano metals. I tested silicon-memristors after mechanical fracture, after pressure treatment in a liquid-filled pressure vessel (in our own lab, and in Takasago Works, Japan), and after bombarded by a 22 GeV electron bunch (at the FACET facility in Stanford Linear Accelerator Center) that created a magnetic pressure. I also tested them under a magnetic field in a cryostat of millikelvin temperature (at NHMFL, Florida). The results revealed that the memristor has a random three-dimensional pathway on which electrons can diffuse, and like other amorphous networks it has soft spots at which electrons are trapped, forming floating gates that regulate the diffusion of untrapped electrons. Indeed, pure amorphous nano silicon conducts in the same way even though it does not switch.
Silicon-memristor is thus fundamentally different from the prevailing ionic memristors in two fundamental ways. First, it is a purely electronic memristor. Second, it has a uniform set of conducting pathways. This is unlike ionic memristors that rely on either filaments or Schottky barriers at electrode/film interfaces (many if not all also have nano filaments.) Uniform, nanometallic switching in silicon-memristor is unprecedentedly stable, fast and power-efficient. It could be readily integrated into silicon technology for data storage, in-memory computation, and biology-inspired computing.
The related paper has been published in Nature Electronics. See details:
Yang Lu, Ana Alvarez, Chung-Ho Kao, Jong-Shing Bow, San-Yuan Chen, I-Wei Chen, An electronic silicon-based memristor with a high switching uniformity, Nat Electron, 2(2), 66, 2019.