Plasma etching effects, such as microtrenching and bowing, negatively impact device performance. Modeling of these effects at nanoscale is challenging, and theoretical and experimental investigations are highly desired to gain insights into mechanisms. In this paper, we propose a new plasma etching model based on Monte Carlo simulations with a cellular method. This model considers reactions and ion-enhanced etching and consists of a novel particle reflection algorithm, which is a key factor impacting the etch profile. This model reproduces the adjustable microtrenching and bowing effects in periodic dense trenches with tens of nanometer dimensions. We conduct experiments of Si etching by Cl2 and validate the model by comparing the simulated profile with cross-sectional scanning electron microscope images. This work enables a potential physical model driven process emulation tool toward design technology co-optimization.
Gate-all-around (GAA) structures are important for future logic devices and 3D-DRAM. Inner-spacer cavity etching and channel release both require selective etching of Si0.7Ge0.3. Increasing the number of channel-stacking layers is an effective way to improve device current-driving capability and storage density. Previous work investigated ICP selective etching of a three-cycle Si0.7Ge0.3/Si multilayer structure and the related etching effects. This study focuses on the dry etching of a 15-cycle Si0.7Ge0.3/Si multilayer structure and the associated etching effects, using simulation and experimentation. The simulation predicts the random effect of lateral etching depth and the asymmetric effect of silicon nanosheet damage on the edge, both of which are verified by experiments. Furthermore, the study experimentally investigates the influence and mechanism of pressure, power, and other parameters on the etching results. Research on these etching effects and mechanisms will provide important points of reference for the dry selective etching of Si0.7Ge0.3 in GAA structures.
Machine learning has advanced unprecedentedly, exemplified by GPT-4 and SORA. However, they cannot parallel human brains in efficiency and adaptability due to differences in signal representation, optimization, runtime reconfigurability, and hardware architecture. To address these challenges, we introduce pruning optimization for input-aware dynamic memristive spiking neural network (PRIME). PRIME uses spiking neurons to emulate brain’s spiking mechanisms and optimizes the topology of random memristive SNNs inspired by structural plasticity, effectively mitigating memristor programming stochasticity. It also uses the input-aware early-stop policy to reduce latency and leverages memristive in-memory computing to mitigate von Neumann bottleneck. Validated on a 40-nm, 256-K memristor-based macro, PRIME achieves comparable classification accuracy and inception score to software baselines, with energy efficiency improvements of 37.8× and 62.5×. In addition, it reduces computational loads by 77 and 12.5% with minimal performance degradation and demonstrates robustness to stochastic memristor noise. PRIME paves the way for brain-inspired neuromorphic computing.
18 January 2022 in IEEE Transactions on Semiconductor Manufacturing
Silicon oxynitride (SiO x N y ) is a common barrier material in thin-film encapsulation (TFE) organic light-emitting diode (OLED). Substrate defects, voids and film internal defects occur during SiO x N y deposition process and result in poor film conformity and barrier failure. In this work, a mathematical model is built to evaluate experimental deposition rate and a high accuracy two dimensional model is proposed to predict the SiO x N y thin film profile evolution on arbitrary substrate in plasma enhanced chemical vapor deposition (PECVD) process. Based on level set method, image processing algorithm, iterative algorithm and velocity extension algorithm are proposed to ensure the high accuracy simulation. A series of design of experiments (DOEs) of SiO x N y film deposition are conducted to validate the model. For numerical deposition rate, the model predicted value fits the experimental data quite well and the offset in between has a root-mean-square error of 2.33%. For film’s cross section profile, satisfactory agreement between model and scanning electron microscope (SEM) image is obtained, we use the step coverage as quantitative index for film uniformity, and the index’s prediction errors are all less than 4%.
Profile non-idealities after plasma etching negatively affect the performance of advanced node devices. The charging effect is a well-known mechanism affecting the hardmask and substrate profiles. Unfortunately, directly characterizing this phenomenon in practical processes is extremely challenging. In this article, we propose a novel etching model to overcome this challenge by incorporating algorithms that simultaneously account for charging effects and particle reflection mechanisms. The model is able to reproduce the real-time profile evolution of both hardmask and substrate layers in an advanced nanoscale etching process. We calculate the electric field distribution induced by the surface charges accumulated on the hardmask, which affects both the trajectory of individual incident ions and the overall etching profiles. To validate our approach, we perform experiments of Si etching in Cl2 plasma and compare the simulated profiles with scanning electron microscope images. The model also identifies the impact of the charging effect on profile defects, such as mask faceting, substrate bowing, and microtrenching. This work provides insights into the charging effect mechanism and its influence on the etching profile, ultimately providing new knobs for advanced process development and optimization.
1 January 2018 in Journal of Microelectronic Manufacturing