Neural networks are an increasingly attractive algorithm for natural language processing and pattern recognition. Deep networks with >50 M parameters are made possible by modern graphics processing ...unit clusters operating at <50 pJ per op and more recently, production accelerators are capable of <5 pJ per operation at the board level. However, with the slowing of CMOS scaling, new paradigms will be required to achieve the next several orders of magnitude in performance per watt gains. Using an analog resistive memory (ReRAM) crossbar to perform key matrix operations in an accelerator is an attractive option. This paper presents a detailed design using the state-of-the-art 14/16 nm process development kit for of an analog crossbar circuit block designed to process three key kernels required in training and inference of neural networks. A detailed circuit and device-level analysis of energy, latency, area, and accuracy are given and compared with relevant designs using standard digital ReRAM and static random access memory (SRAM) operations. It is shown that the analog accelerator has <inline-formula> <tex-math notation="LaTeX">270\times </tex-math></inline-formula> energy and <inline-formula> <tex-math notation="LaTeX">540\times </tex-math></inline-formula> latency advantage over a similar block utilizing only digital ReRAM and takes only 11 fJ per multiply and accumulate. Compared with an SRAM-based accelerator, the energy is <inline-formula> <tex-math notation="LaTeX">430\times </tex-math></inline-formula> better and latency is <inline-formula> <tex-math notation="LaTeX">34\times </tex-math></inline-formula> better. Although training accuracy is degraded in the analog accelerator, several options to improve this are presented. The possible gains over a similar digital-only version of this accelerator block suggest that continued optimization of analog resistive memories is valuable. This detailed circuit and device analysis of a training accelerator may serve as a foundation for further architecture-level studies.
This paper presents a memristive device model capable of accurately matching a wide range of characterization data collected from a tantalum oxide memristor. Memristor models commonly use a set of ...equations and fitting parameters to match the complex dynamic conductivity pattern observed in these devices. Along with the proposed model, a procedure is also described that can be used to optimize each fitting parameter in the model relative to an I-V curve. Therefore, model parameters are self-updated based on this procedure when a new cyclic I-V sweep is provided for model optimization. This model will automatically provide the best possible match to the characterization data without any additional optimization from the user. In this paper, multiple cyclic I-V characterizations are modeled from ten different tantalum oxide devices (on the same wafer). Additionally, studies were completed to demonstrate the amount of variation present between devices on a wafer, as well as the amount of variation present within a single device. Methods for modeling this variation are then proposed, resulting in an accurate and complete, automated, memristor modeling approach.
Charge trapping and slow (from 10 s to >; 1000 s) detrapping in AlGaN/GaN high electron mobility transistors (HEMTs) designed for high breakdown voltages ( >; 1500 V) is studied through a combination ...of electrical, thermal, and optical methods to identify the impact of Al molefraction and passivation on trapping. Trapping due to 5-10 V drain bias stress in the on-state (V gs = 0) is found to have significantly slower recovery, compared with trapping in the off-state (V gs <; V th , V ds = 0). Two different trapping components, i.e., TG1 (E a = 0.6 eV) and TG2 (with negligible temperature dependence), in AlGaN dominate under gate bias stress in the off-state. Al 0.15 Ga 0.85 N shows much more vulnerability to trapping under gate stress in the absence of passivation than does AlGaN with a higher Al mole fraction. Under large drain bias, trapping is dominated by a much deeper trap TD. Detrapping under monochromatic light shows TD to have E a 1.65 eV. Carbon doping in the buffer is shown to introduce threshold voltage shifts, unlike any of the other traps.
Experimental results show the response of 22 nm fully depleted silicon on insulator (FDSOI) transistors to 10 keV x-rays. The extracted coupling factor between the front- and back-gates of the device ...shows a divergence from the traditionally assumed coupling factor exhibiting a non-linear dependence on total ionizing dose (TID). A depletion region in the well under the buried oxide is included in the model of the device. The coupling factor is rederived using the new model following the previous derivation. The new coupling factor has a dependence on TID received by the device. Verification of the new model is performed with comparisons to the experimental data and technology computer-aided design (TCAD) simulations. The new model fits well to data and simulations.
The steady‐state solution of filamentary memristive switching may be derived directly from the heat equation, modelling vertical and radial heat flow. This solution is shown to provide a continuous ...and accurate description of the evolution of the filament radius, composition, heat flow, and temperature during switching, and is shown to apply to a large range of switching materials and experimental time‐scales.
The total ionizing dose response of 128 distinct conductance states in 40 nm SONOS charge-trap memory was experimentally characterized, which reveals in fine detail the analog state dependence of the ...ionizing radiation effect. The 128 states were programmed onto an array of 128K SONOS cells that were irradiated by Co-60 gamma rays up to 1.5 Mrad(Si) total dose. The observed response after radiation and subsequent annealing suggests that radiation deposits a net positive charge in traps that have both shallow and deep energy levels within the nitride bandgap. We use the measured data to simulate a SONOS-based analog in-memory computing accelerator operating under radiation, and evaluate its accuracy on large image recognition neural networks. The impact of ionizing radiation on the algorithm depends on the regime of operation of the irradiated SONOS devices, with states in the weak inversion regime having a larger effect on accuracy. Periodic refreshes of the SONOS states are expected to enable reliable, efficient, and long-term operation in space.