Human–machine collaboration for improving semiconductor process development

Abstract

One of the bottlenecks to building semiconductor chips is the increasing cost required to develop chemical plasma processes that form the transistors and memory storage cells1,2. These processes are still developed manually using highly trained engineers searching for a combination of tool parameters that produces an acceptable result on the silicon wafer3. The challenge for computer algorithms is the availability of limited experimental data owing to the high cost of acquisition, making it difficult to form a predictive model with accuracy to the atomic scale. Here we study Bayesian optimization algorithms to investigate how artificial intelligence (AI) might decrease the cost of developing complex semiconductor chip processes. In particular, we create a controlled virtual process game to systematically benchmark the performance of humans and computers for the design of a semiconductor fabrication process. We find that human engineers excel in the early stages of development, whereas the algorithms are far more cost-efficient near the tight tolerances of the target. Furthermore, we show that a strategy using both human designers with high expertise and algorithms in a human first–computer last strategy can reduce the cost-to-target by half compared with only human designers. Finally, we highlight cultural challenges in partnering humans with computers that need to be addressed when introducing artificial intelligence in developing semiconductor processes.

Multi-fidelity Bayesian optimization of covalent organic frameworks for xenon/krypton separations

“…BO has been deployed for the optimization and discovery of many different materials 12,[23][24][25][26] in the lab or a computer simulation, including nanoporous materials, [27][28][29][30][31] nanoparticles, 32 light emitting diodes, 33 carbon nanotubes, 34 photovoltaics, [35][36][37] additively manufactured structures, 38 polymers, [39][40][41][42][43] thermoelectrics, 44 anti-microbial active surfaces, 45 quantum dots, 46 luminescent materials, 47 catalysts, [48][49][50][51][52] thin lms, 53 solid chemical propellants, 54 alloys, 55 and phase-change memory materials. 56 More, BO has been used to optimize processes to synthesize materials and chemicals [57][58][59][60][61][62][63] or to employ materials for an industrial-scale task. 64 Multi-delity Bayesian optimization for materials discovery Oen, we have multiple options of different experiments to measure/evaluate/predict the relevant property of the material-experiments that trade (1) delity, i.e.…”

“…BO has been deployed for the optimization and discovery of many different materials 12,23–26 in the lab or a computer simulation, including nanoporous materials, 27–31 nanoparticles, 32 light emitting diodes, 33 carbon nanotubes, 34 photovoltaics, 35–37 additively manufactured structures, 38 polymers, 39–43 thermoelectrics, 44 anti-microbial active surfaces, 45 quantum dots, 46 luminescent materials, 47 catalysts, 48–52 thin films, 53 solid chemical propellants, 54 alloys, 55 and phase-change memory materials. 56 More, BO has been used to optimize processes to synthesize materials and chemicals 57–63 or to employ materials for an industrial-scale task. 64…”

Review: Machine learning for advancing low-temperature plasma modeling and simulation

“…In contrast to real time process control and estimation, a study by Kanarik et al 152 has investigated ML aided process recipe design. A virtual process of a radio-frequency etch plasma with fluorocarbon and O 2 chemistry coupled to a feature profile simulator to estimate etching and critical dimension (CD) has been used for reference.…”

“…In contrast to real time process control and estimation, a study by Kanarik et al 152 . has investigated ML aided process recipe design.…”

Driving school for self-driving labs

“…Indeed, there have been recent reports highlighting the importance of the human–machine collaboration and how this collaboration can be used as part of otherwise autonomous systems. 28,29 However, given that we are only now seeing the widespread adoption of SDLs, there do not yet exist resources to help experimenters know what to monitor and what to adjust.…”