-
Model Performance Prediction for Hyperparameter Optimization of Deep Learning Models Using High Performance Computing and Quantum Annealing
Authors:
Juan Pablo García Amboage,
Eric Wulff,
Maria Girone,
Tomás F. Pena
Abstract:
Hyperparameter Optimization (HPO) of Deep Learning-based models tends to be a compute resource intensive process as it usually requires to train the target model with many different hyperparameter configurations. We show that integrating model performance prediction with early stop** methods holds great potential to speed up the HPO process of deep learning models. Moreover, we propose a novel a…
▽ More
Hyperparameter Optimization (HPO) of Deep Learning-based models tends to be a compute resource intensive process as it usually requires to train the target model with many different hyperparameter configurations. We show that integrating model performance prediction with early stop** methods holds great potential to speed up the HPO process of deep learning models. Moreover, we propose a novel algorithm called Swift-Hyperband that can use either classical or quantum support vector regression for performance prediction and benefit from distributed High Performance Computing environments. This algorithm is tested not only for the Machine-Learned Particle Flow model used in High Energy Physics, but also for a wider range of target models from domains such as computer vision and natural language processing. Swift-Hyperband is shown to find comparable (or better) hyperparameters as well as using less computational resources in all test cases.
△ Less
Submitted 29 November, 2023;
originally announced November 2023.
-
Ultrasonic Delamination Based Adhesion Testing for High-Throughput Assembly of van der Waals Heterostructures
Authors:
Tara Peña,
Jewel Holt,
Arfan Sewaket,
Stephen M. Wu
Abstract:
Two-dimensional (2D) materials assembled into van der Waals (vdW) heterostructures contain unlimited combinations of mechanical, optical, and electrical properties that can be harnessed for potential device applications. Critically, these structures require control over interfacial adhesion in order for them to be constructed and to have enough integrity to survive industrial fabrication processes…
▽ More
Two-dimensional (2D) materials assembled into van der Waals (vdW) heterostructures contain unlimited combinations of mechanical, optical, and electrical properties that can be harnessed for potential device applications. Critically, these structures require control over interfacial adhesion in order for them to be constructed and to have enough integrity to survive industrial fabrication processes upon their integration. Here, we promptly determine the quality of adhesion of various exfoliated 2D materials on conventional SiO$_{2}$/Si substrates using ultrasonic delamination threshold testing. This test allows us to quickly infer relative substrate adhesion based on the percent area of 2D flakes that survive a fixed time in an ultrasonic bath, allowing for control over process parameters that yield high or poor adhesion. We leverage this control of adhesion to optimize the vdW heterostructure assembly process, where we show that samples with high or low substrate adhesion relative to each other can be used selectively to construct high-throughput vdW stacks. Instead of tuning the adhesion of polymer stamps to 2D materials with constant 2D-substrate adhesion, we tune the 2D-substrate adhesion with constant stamp adhesion to 2D materials. The polymer stamps may be reused without any polymer melting steps, thus avoiding high temperatures (<120°C) and allowing for high-throughput production. We show that this procedure can be used to create high-quality 2D twisted bilayer graphene on SiO$_{2}$/Si, characterized with atomic force microscopy and Raman spectroscopic map**, as well as low-angle twisted bilayer WSe$_{2}$ on h-BN/SiO$_{2}$/Si, where we show direct real-space visualization of moiré reconstruction with tilt angle-dependent scanning electron microscopy.
△ Less
Submitted 9 December, 2022; v1 submitted 7 October, 2022;
originally announced October 2022.
-
Moiré Engineering in 2D Heterostructures with Process-Induced Strain
Authors:
Tara Peña,
Aditya Dey,
Shoieb A. Chowdhury,
Ahmad Azizimanesh,
Wenhui Hou,
Arfan Sewaket,
Carla L. Watson,
Hesam Askari,
Stephen M. Wu
Abstract:
We report deterministic control over moiré superlattice interference pattern in twisted bilayer graphene by implementing designable device-level heterostrain with process-induced strain engineering, a widely used technique in industrial silicon nanofabrication processes. By depositing stressed thin films onto our twisted bilayer graphene samples, heterostrain magnitude and strain directionality ca…
▽ More
We report deterministic control over moiré superlattice interference pattern in twisted bilayer graphene by implementing designable device-level heterostrain with process-induced strain engineering, a widely used technique in industrial silicon nanofabrication processes. By depositing stressed thin films onto our twisted bilayer graphene samples, heterostrain magnitude and strain directionality can be controlled by stressor film force (film stress x film thickness) and patterned stressor geometry, respectively. We examine strain and moiré interference with Raman spectroscopy through in-plane and moiré-activated phonon mode shifts. Results support systematic C$_{3}$ rotational symmetry breaking and tunable periodicity in moiré superlattices under the application of uniaxial or biaxial heterostrain. Experimental results are validated by molecular statics simulations and density functional theory based first principles calculations. This provides a method to not only tune moiré interference without additional twisting, but also allows for a systematic pathway to explore different van der Waals based moiré superlattice symmetries by deterministic design.
△ Less
Submitted 3 April, 2023; v1 submitted 7 October, 2022;
originally announced October 2022.
-
Strain-Based Room-Temperature Non-Volatile MoTe$_2$ Ferroelectric Phase Change Transistor
Authors:
Wenhui Hou,
Ahmad Azizimanesh,
Arfan Sewaket,
Tara Peña,
Carla Watson,
Ming Liu,
Hesam Askari,
Stephen M. Wu
Abstract:
The primary mechanism of operation of almost all transistors today relies on electric-field effect in a semiconducting channel to tune its conductivity from the conducting 'on'-state to a non-conducting 'off'-state. As transistors continue to scale down to increase computational performance, physical limitations from nanoscale field-effect operation begin to cause undesirable current leakage that…
▽ More
The primary mechanism of operation of almost all transistors today relies on electric-field effect in a semiconducting channel to tune its conductivity from the conducting 'on'-state to a non-conducting 'off'-state. As transistors continue to scale down to increase computational performance, physical limitations from nanoscale field-effect operation begin to cause undesirable current leakage that is detrimental to the continued advancement of computing. Using a fundamentally different mechanism of operation, we show that through nanoscale strain engineering with thin films and ferroelectrics (FEs) the transition metal dichalcogenide (TMDC) MoTe$_2$ can be reversibly switched with electric-field induced strain between the 1T'-MoTe$_2$ (semimetallic) phase to a semiconducting MoTe$_2$ phase in a field effect transistor geometry. This alternative mechanism for transistor switching sidesteps all the static and dynamic power consumption problems in conventional field-effect transistors (FETs). Using strain, we achieve large non-volatile changes in channel conductivity (G$_{on}$/G$_{off}$~10$^7$ vs. G$_{on}$/G$_{off}$~0.04 in the control device) at room temperature. Ferroelectric devices offer the potential to reach sub-ns nonvolatile strain switching at the attojoule/bit level, having immediate applications in ultra-fast low-power non-volatile logic and memory while also transforming the landscape of computational architectures since conventional power, speed, and volatility considerations for microelectronics may no longer exist.
△ Less
Submitted 17 May, 2019;
originally announced May 2019.