Search

Ferroelectric (FE) materials are being studied for a variety of applications in memory, logic, and neuromorphic computing, for which predictive models of FE polarization are essential. In this paper, we present a Monte Carlo simulation framework capable of predicting the dynamic, history-dependent response of an FE under arbitrary input waveforms. The simulation is developed by generalizing the physics-based nucleation-limited switching model for polarization reversal in a polycrystalline FE. Measured polarization reversal data from fabricated FE Hf0.5Zr0.5O2 capacitors are used to extract the statistical distribution of FE grains. After parameter extraction, the model is able to predict the dynamics of the FE capacitor without further calibration. Finally, the model is applied to characterize the dynamic response of FE–dielectric bilayer structures and quantify the reduction in memory window due to device variability.

Ferroelectric Zr-doped HfO2 (HZO) is a promising candidate for steep slope transistors and memory technology. For these applications, it is essential to understand and optimize the switching dynamics of the ferroelectric film. In this letter, we characterize the polarization reversal of an 8 nm-thick HZO film deposited by the atomic layer deposition with voltage pulses varying in amplitude (0.8–2 V) and duration (200 ns–7.6 ms). We show that the measurements are well described by a nucleation limited switching model, which enables extraction of the minimum switching time and the probability distribution of local electric field variations in the polycrystalline film. The close model fit spanning 5 orders of magnitude in pulse duration indicates the applicability of this model to HZO. This characterization framework can be used to quantify, compare, and optimize the switching dynamics of ferroelectric HZO.

A back-gated multilayer nanoribbon molybdenum disulfide (MoS2) transistor grown by chemical vapor transport and doped using polyethylene oxide cesium perchlorate is fabricated and characterized. Ions in the polymer dielectric are directed by side gates to the source and drain access regions where they form electric double layers (EDLs) that control the carrier densities. This allows the junctions of the same transistor channel to be reconfigured as an n-MOSFET, p-MOSFET, and as a tunnel field-effect transistors. The EDLs are formed at room temperature and then locked into place by cooling the polymer below the glass transition temperature (~240 K). Transport measurements are presented and explained using simulated band diagrams. Both n and p-conduction in MoS2 is demonstrated using solid polymer ion doping, enabling characterization of a semiconductor in which the doping of the same channel has been reconfigured to form three different transistor configurations.

We propose a technology-independent method, referred to as adjacent connection matrix (ACM), to efficiently map signed weight matrices to non-negative crossbar arrays. When compared to same-hardware-overhead mapping methods, using ACM leads to improvements of up to 20% in training accuracy for ResNet-20 with the CIFAR-10 dataset when training with 5-bit precision crossbar arrays or lower. When compared with strategies that use two elements to represent a weight, ACM achieves comparable training accuracies, while also offering area and read energy reductions of 2.3x and 7x, respectively. ACM also has a mild regularization effect that improves inference accuracy in crossbar arrays without any retraining or costly device/variation-aware training., Comment: Accepted at DAC'20

Complex gate dielectrics are being widely explored to provide voltage amplification leading to steep subthreshold swing (SS) in transistors [1]. Here we show that hysteretic steep SS can arise simply from the introduction of a series resistance in the gate of a metal-oxide-semiconductor field-effect transistor (MOSFET), i.e. as a dynamic effect. The choice of measurement slew rate (SR) and slew direction then directly affects the measured SS as has been previously reported [2]. We use an ionic polymer gate dielectric to illustrate how in this case the series resistance arises from the ionic conductance, which controls the time response of the electric double layer (EDL) at the polymer/metal [3] or polymer/semiconductor interface [4]. While the ionic polymer in an EDL transistor is used to illustrate the effects, the principles apply generally to other complex insulating gates including ferroelectrics and piezoelectrics.

A physics-based, circuit-compatible Monte Carlo simulation framework, capable of predicting the dynamic response of a ferroelectric (FE) under any arbitrary input waveform, is developed by extending the nucleation-limited switching model. Measured polarization reversal data from fabricated FE W/Hf 0.5 Zr 0.5 O 2 (HZO)/W capacitors is used to extract the statistical distribution of FE grains, which show negligible variation with film thickness. After parameter extraction, the model is able to predict the dynamics of HZO and bilayer HZO/HfO 2 (FE-DE) thin films without further calibration. Unlike prior models, the proposed model is able to predict device-to-device variability, and quantify the resultant reduction in the memory window for highly scaled devices, revealing a significant reduction for FE capacitors having $\sim 40\times 40\ \text{nm}^{2}$ ). The memory window is further reduced in FE-DE stacks for the same programming voltage and pulse duration due to the dielectric depolarizing field.

Based on measurements of a vertical nanowire InAs/GaAsSb/GaSb tunneling field-effect transistor (TFET) that exhibited minimum subthreshold swing of 48 mV/dec and a record high I 60 of 0.31 μA/μm, a SPICE model has been generated to allow an experimentally-based prediction of the nanowire TFET technology. At 30 GHz the detector has been simulated to reveal a sensitivity of 4.8 kV/W biased near zero volts (V GS = −0.06 V, V DS = 0.1 V). A maximum sensitivity of over 4000 kV/W has been obtained under biased conditions. These results exceed prior measurements of an In 0 . 53 Ga 0 . 47 As/ GaAs 0 . 5 Sb 0 . 5 heterojunction TFET by over an order of magnitude.

Gokmen and Vlasov proposed that the training of deep neural networks would be dramatically accelerated by the realization of resistive processing units that can store analog weights to minimize data movement during training [1]. Here we measure and evaluate the partial switching of the ferroelectric (FE) PbZrTiO 3 (PZT) and, for the first time, Hf 0 8 Zr 0 2 O 2 , (HZO), formed by atomic layer deposition, for use as a nonvolatile analog memory element. The spontaneous polarization P of a FE can be switched by applying a voltage that exceeds the coercive voltage for switching of the FE. This polarization, when fully switched, can have a value of +P or −P (C/cm2), depending on the polarity of the applied voltage. Ferroelectric memory is typically operated in fully switched +P or −P polarizations, however with short (< 10 μs) pulsed voltages, the FE can be partially switched, i.e. only a fraction of the domains are switched during the pulse duration as shown by Tokumitsu in PZT [2] and indicated in Fig. 1.

Low voltage transistors are being developed to achieve steep, less than 60 mV/decade, subthreshold swings at room temperature. This paper outlines progress, technical challenges, and applications for these devices.

Digital correlated double sampling (DCDS), a readout technique for charge-coupled devices (CCD), is gaining popularity in astronomical applications. By using an oversampling ADC and a digital filter, a DCDS system can achieve a better performance than traditional analogue readout techniques at the expense of a more complex system analysis. Several attempts to analyse and optimize a DCDS system have been reported, but most of the work presented in the literature has been experimental. Some approximate analytical tools have been presented for independent parameters of the system, but the overall performance and trade-offs have not been yet modelled. Furthermore, there is disagreement among experimental results that cannot be explained by the analytical tools available. In this work, a theoretical analysis of a generic DCDS readout system is presented, including key aspects such as the signal conditioning stage, the ADC resolution, the sampling frequency and the digital filter implementation. By using a time-domain noise model, the effect of the digital filter is properly modelled as a discrete-time process, thus avoiding the imprecision of continuous-time approximations that have been used so far. As a result, an accurate, closed-form expression for the signal-to-noise ratio at the output of the readout system is reached. This expression can be easily optimized in order to meet a set of specifications for a given CCD, thus providing a systematic design methodology for an optimal readout system. Simulated results are presented to validate the theory, obtained with both time- and frequency-domain noise generation models for completeness., This is a pre-copyedited, author-produced PDF of an article accepted for publication in Monthly Notices of the Royal Astronomical Society following peer review. The version of record is available online via the DOI or at: http://mnras.oxfordjournals.org/content/455/2/1443.full.pdf