Reconfigurable Multifunctional van der Waals Ferroelectric Devices and Logic Circuits

One transistor. Simultaneously a p-type device, an n-type device, an ambipolar device, a diode, a photodetector, and a logic gate — depending entirely on what you wrote into its memory last. Not different transistors swapped out on a board, but the same physical device, reconfigured in place. What kind of transistor holds all of that at once? The answer, from Ankita Ram and colleagues, is one built from two-atom-thick crystals sitting on a ferroelectric gate that remembers. Modern integrated circuits are running into hard walls. Conventional transistors are fixed at the factory — a p-type device is p-type forever, baked in by chemical doping and fabrication choices. As dimensions shrink, that inflexibility drives up circuit complexity and energy costs. Then there's the von Neumann bottleneck: computation and memory are physically separated, so data must be constantly shuttled between storage and processor. That shuttle costs energy and limits throughput. Logic-in-memory is the proposed alternative — computation happening where the data already lives, in a single device that stores information and participates in logic operations simultaneously. Ram and colleagues describe this as a paradigm shift, and their reconfigurable ferroelectric field-effect transistor, or Re-FeFET, is their demonstration of how to get there. Two earlier ideas point toward this device. Reconfigurable field-effect transistors, or ReFETs, use two independent gates — one sets the carrier type, while the other controls switching — so a single device can be toggled between n-type and p-type at runtime. Ferroelectric field-effect transistors, or FeFETs, use a ferroelectric gate that retains a polarization state without continuous power, electrostatically doping the adjacent semiconductor so the device remembers its configuration. The Re-FeFET combines both: a reconfigurable gate geometry with a ferroelectric gate material that makes the configuration non-volatile. The physical stack is elegant. The semiconductor channel is tungsten diselenide, or WSe2 — a few nanometers thick, exfoliated from bulk crystal. Beneath it sits a thin layer of hexagonal boron nitride, known as h-BN, acting as a dielectric spacer to improve transport and reduce leakage. Beneath that is the ferroelectric gate material: copper indium thiophosphate, or CuInP2S6, abbreviated as CIPS. All three are van der Waals materials — meaning they're held together by weak surface forces rather than chemical bonds, so they can be peeled into atomically thin flakes and stacked layer by layer without conventional epitaxy. The result is a clean, sharp interface between materials that would be nearly impossible to combine any other way. The crucial architectural choice is the split gate: two independent bottom electrodes beneath different portions of the same WSe2 channel. One is the Program Gate, which sets the ferroelectric polarization and therefore the doping state of the channel beneath it. The other is the Control Gate, which handles the actual switching — turning the device on and off during operation. Typical channel lengths are around six micrometers. When the Program Gate is pulsed to plus twelve volts, the ferroelectric polarizes upward, accumulating electrons in the WSe2 below — n-type. Pulse it to minus twelve volts, the polarization flips, holes accumulate — p-type. Because ferroelectric polarization is remanent, that doping state persists with no power applied. Two independently programmable gates beneath one channel produce four distinct homojunction profiles: p-p, n-n, p-n, and n-p. That's where the four personalities come from. In the p-p or n-n configurations, the device behaves as a unipolar transistor. The p-type program state achieves an on-off current ratio exceeding one million — a factor of a million between the conducting and non-conducting states. The n-type state reaches ten thousand. The ferroelectric memory window — the voltage span over which the remanent polarization stores a distinct state — approaches nearly ten volts. That window is what makes the device non-volatile: it doesn't need to be held in place by an external bias. Switch to a p-n or n-p configuration, and the device becomes a diode. The asymmetric junction produces strongly nonlinear current-voltage curves, with a rectification ratio — forward current divided by reverse current — exceeding ten thousand at one volt. The p-p and n-n states, by contrast, show near-linear behavior with rectification ratios barely above one. That's the difference between a diode and a wire, produced by nothing more than a programming pulse. And in the diode state, something else happens. The built-in electric field of the ferroelectrically defined p-n junction separates photogenerated electron-hole pairs — the electron goes one way, the hole goes the other — generating a measurable photocurrent and photovoltage. In the p-n configuration, the short-circuit photocurrent reaches 1.2 nanoamps and the open-circuit photovoltage reaches 0.7 volts. Flip to the n-p configuration, and both reverse sign: minus 1.3 nanoamps, minus 0.7 volts. The same device that computes can harvest light, and it switches between those roles with a gate pulse. This is where the logic story begins in earnest. Ram and colleagues abstract the Re-FeFET into a parameter set: t, the non-volatile programmed type — zero for p and one for n — and X, the non-volatile ferroelectric polarization of the control-gate sub-layer. Against those stored states, the volatile gate voltages drive switching during operation. The team maps out a conduction table — which combinations of programmed state and applied voltage result in the device's on or off state — and uses that table as a design tool for building logic circuits. Two operating modes emerge. In electronic mode, both logic inputs are volatile gate voltages: the Control Gate and Program Gate are driven to represent binary one or zero. A single Re-FeFET plus a load resistor experimentally implements an XNOR function, confirmed by waveform measurements at a supply voltage of one volt. In ferroelectric mode, one input is the remanent polarization state itself — which of the two ferroelectric branches the device is currently sitting on — and the other is a volatile gate voltage. Depending on how the Program Gate is pre-set, the same physical series circuit produces either a ferroelectric NAND or a ferroelectric AND. The logic function is determined by what was written into the ferroelectric memory, not by any physical change to the circuit. This is polymorphic logic: one circuit, multiple truth tables, selected by programming rather than fabrication. Two series-connected Re-FeFETs can implement all six elementary two-input logic gates — NOT, NAND, NOR, AND, OR, and XOR — by choosing the appropriate entries from the conduction table. The programmed states are genuinely non-volatile. Retention measurements show the four homojunction states holding stable for more than ten thousand seconds after the programming voltage is removed. The circuit-level payoff is where the numbers land hardest. A NAND or NOR gate takes two Re-FeFETs, while the equivalent complementary metal-oxide-semiconductor circuit needs four transistors. An XOR — typically one of the most expensive gates in complementary metal-oxide-semiconductor, requiring around twelve transistors including inverters — is implemented with two Re-FeFETs. That's a factor of six reduction. Across the logic functions explored, the Re-FeFET approach yields up to an eighty percent reduction in transistor count compared to standard complementary metal-oxide-semiconductor. Transistor count is not an abstract metric: it directly determines chip area, power consumption, and heat. Ram and colleagues also note a hardware security dimension. When a circuit's function is determined by a non-volatile internal state rather than its physical layout, reverse-engineering the chip becomes harder — the schematic doesn't tell you what the chip does. And because identical Re-FeFET circuit elements can implement different functions, their power signatures during operation look the same regardless of which logic function is running, complicating side-channel analysis. The authors are candid: reconfigurable ferroelectric logic-in-memory is, in their words, mostly uncharted territory, and this is a proof-of-concept. The device metrics are striking — on-off ratios reaching one million, hysteresis windows up to ten volts, rectification ratios of ten thousand, retention beyond ten thousand seconds — but the path from a laboratory-fabricated van der Waals stack to manufacturable technology involves challenges the paper does not claim to have solved. What it does establish is that two-dimensional materials can be engineered into computing primitives that simultaneously store, switch, rectify, and detect light, all within a single device whose function is written in ferroelectric memory and can be rewritten at runtime. That's a different picture of what a transistor can be. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.