arXiv subjects

Charles Evans

Publications and source records attributed to Charles Evans.

3 recordsLinked to original sources

arXiv2024

Low Leakage Ferroelectric Heteroepitaxial Al$_{0.7}$Sc$_{0.3}$N Films on GaN

Wurtzite (Al,Sc)N ferroelectrics are attractive for microelectronics applications due to their chemical and epitaxial structural compatibility with wurtzite semiconductors such as GaN and (Al,Ga)N. However, the leakage current in epitaxial stacks reported to date should be reduced for reliable device operation. Following the tradition of other semiconductor heterostructures, crystalline structural quality, as measured by breadth of diffraction peaks and correlating with dislocation density, is commonly used as a proxy for leakage current, but we demonstrate here that the crystalline mosaicity that dominates the broadening of diffraction peaks in epitaxial Al$_{0.7}$Sc$_{0.3}$N stacks does not dominate leakage current. We report here well-saturated ferroelectric hysteresis loops and orders of magnitude lower leakage current (0.07 A cm$^{-2}$) compared to values reported in literature (1 ~ 19 A cm$^{-2}$) for sputter-deposited epitaxial Al$_{0.7}$Sc$_{0.3}$N/GaN of comparable crystalline quality to prior reports. Further, we show Al$_{0.7}$Sc$_{0.3}$N on lattice-matched InGaN buffers with improved structural characteristics exhibits increased leakage characteristics. This demonstration and understanding can help to guide further efforts towards reliable wurtzite ferroelectric devices and prioritize approaches targeting further leakage current reduction.

physics.app-ph↗

arXiv2023

User Tampering in Reinforcement Learning Recommender Systems

In this paper, we introduce new formal methods and provide empirical evidence to highlight a unique safety concern prevalent in reinforcement learning (RL)-based recommendation algorithms -- 'user tampering.' User tampering is a situation where an RL-based recommender system may manipulate a media user's opinions through its suggestions as part of a policy to maximize long-term user engagement. We use formal techniques from causal modeling to critically analyze prevailing solutions proposed in the literature for implementing scalable RL-based recommendation systems, and we observe that these methods do not adequately prevent user tampering. Moreover, we evaluate existing mitigation strategies for reward tampering issues, and show that these methods are insufficient in addressing the distinct phenomenon of user tampering within the context of recommendations. We further reinforce our findings with a simulation study of an RL-based recommendation system focused on the dissemination of political content. Our study shows that a Q-learning algorithm consistently learns to exploit its opportunities to polarize simulated users with its early recommendations in order to have more consistent success with subsequent recommendations that align with this induced polarization. Our findings emphasize the necessity for developing safer RL-based recommendation systems and suggest that achieving such safety would require a fundamental shift in the design away from the approaches we have seen in the recent literature.

cs.AI↗

arXiv2014

Lorenz gauge gravitational self-force calculations of eccentric binaries using a frequency domain procedure

We present an algorithm for calculating the metric perturbations and gravitational self-force for extreme-mass-ratio inspirals (EMRIs) with eccentric orbits. The massive black hole is taken to be Schwarzschild and metric perturbations are computed in Lorenz gauge. The perturbation equations are solved as coupled systems of ordinary differential equations in the frequency domain. Accurate local behavior of the metric is attained through use of the method of extended homogeneous solutions and mode-sum regularization is used to find the self-force. We focus on calculating the self-force with sufficient accuracy to ensure its error contributions to the phase in a long term orbital evolution will be $δΦ\lesssim 10^{-2}$ radians. This requires the orbit-averaged force to have fractional errors $\lesssim 10^{-8}$ and the oscillatory part of the self-force to have errors $\lesssim 10^{-3}$ (a level frequently easily exceeded). Our code meets this error requirement in the oscillatory part, extending the reach to EMRIs with eccentricities of $e \lesssim 0.8$, if augmented by use of fluxes for the orbit-averaged force, or to eccentricities of $e \lesssim 0.5$ when used as a stand-alone code. Further, we demonstrate accurate calculations up to orbital separations of $a \simeq 100 M$, beyond that required for EMRI models and useful for comparison with post-Newtonian theory. Our principal developments include (1) use of fully constrained field equations, (2) discovery of analytic solutions for even-parity static modes, (3) finding a pre-conditioning technique for outer homogeneous solutions, (4) adaptive use of quad-precision and (5) jump conditions to handle near-static modes, and (6) a hybrid scheme for high eccentricities.

gr-qc↗