"Testing Semiconductor Device Performance 10 Times Faster with Artificial Neural Networks"

GIST Professor Hong Seong-min's Team Enables Time and Cost Savings in Semiconductor Devices

[Asia Economy Reporter Kim Bong-su] A technology has been developed that enables faster and lower-cost performance testing of semiconductor devices through artificial neural networks.

The research team led by Professor Hong Sung-min of the Department of Electrical, Electronics and Computer Engineering at Gwangju Institute of Science and Technology (GIST), together with Professor Choi Jong-hyun from the AI Graduate School, announced on the 24th that they have developed a technology that can perform semiconductor device simulations much faster using artificial neural networks.

Semiconductor device simulation is a technology that predicts the performance of semiconductor devices using computer programs, which can significantly reduce the enormous time and cost required for semiconductor device development.

The research team focused on the fact that most of the time spent performing semiconductor device simulations is used to calculate unnecessary intermediate process answers, and succeeded in shortening the simulation time by nearly 10 times by generating excellent approximate solutions with trained artificial neural networks.

Recently, with the global semiconductor shortage, semiconductor manufacturing technology has received great attention. Since it is especially important to complete semiconductor device technology development in a short time, expectations for semiconductor device simulation are high. Running semiconductor device simulation programs usually takes a lot of time, making it a bottleneck in technology development. Moreover, existing techniques such as parallel computing required enormous computing resources to handle numerous device design candidates.

The research team shortened the simulation time by directly obtaining answers only for the voltage conditions that users want to know. Semiconductor device simulation involves solving nonlinear equations, so it is necessary to have an excellent approximate solution close to the correct answer. However, since it is difficult to know in advance the excellent approximate solution for the voltage condition the user wants to know (around approximately 0.7 V), it is inevitably started from 0 V and the voltage is gradually increased.

To directly obtain the answer for the desired voltage condition, the research team introduced artificial neural networks. This artificial neural network performs supervised learning on existing simulation results and generates the potential profile inside the semiconductor device corresponding to the desired situation. The potential profile refers to the electric potential energy that a unit charge possesses.

Since voltage is applied to semiconductor devices, the potential value differs at each position inside the device. The movement of electrons inside the semiconductor device is influenced by this potential profile, making it the most important physical quantity in semiconductor device simulation. To verify the proposed method, a speed comparison with the existing method was conducted. Compared to the result of setting the simulation control parameters of the existing method to optimal values, a speed improvement of more than 8.4 times was achieved. Since the optimal values of the simulation control parameters cannot be known before performing the simulation directly, the expected speed improvement in actual application is more than 10 times.

Professor Hong Sung-min said, “It is significant that we have confirmed for the first time that the execution time of semiconductor device simulation can be greatly reduced by utilizing artificial neural networks,” and added, “We expect it to be actively used in the development of next-generation semiconductor devices through follow-up research.”

This research result was published online on the 7th of this month in the international academic journal in the semiconductor device field, ‘IEEE Transactions on Electron Devices.’

This content was produced with the assistance of AI translation services.