Microelectronics Is Accelerating Towards the Post-moore Era! Mei Yongfeng's Research Group at Fudan University Integrates DNN and Nanofilm Technology to Accurately Analyze the Angle of Incident Light

"Two-dimensional nanofilms self-assemble into three-dimensional microstructures" is considered an important way to manufacture the next generation of microelectronic devices, which is crucial for the upcoming advanced electronic and optoelectronic applications. However, the formation of the final geometric shape of the two-dimensional nanofilm is affected by etching tracks, chemical reactions, aspect ratios, and other complex factors, making it difficult to improve the yield and finished product rate of self-assembled devices during the manufacturing process, which seriously hinders its real transition from laboratory to industrial application.

In order to accurately predict the morphology and design guidance of pre-strained nano-membrane structures, scientists have designed and developed many analytical and numerical methods over a century.Finite element modeling (FEM) can intuitively simulate and predict the behavior of nanofilms after release.In recent years, it has been used by researchers to study and simulate the self-rolling process of nanofilms. However, judging from the results, most experiments are only satisfied with the formation of specific structures or local analysis, and still lack wide applicability and accurate boundary condition models.

In response to this, the research group of Professor Mei Yongfeng from the Department of Materials Science at Fudan University recently published a research result titled "Multilevel design and construction in nanomembrane rolling for three-dimensional angle-sensitive photodetection" in "Nature Communications".This study proposed a multi-level quasi-static finite element analysis method, and based on this method, designed and constructed six types of silicon (Si) / chromium (Cr) nanofilm assembled three-dimensional microstructures and corresponding three-dimensional photodetectors.The good versatility and industrial practicability of this technology have been fully verified.

Research highlights:

• The multi-level quasi-static finite element method is successfully applied to a wide range of material systems, nanofilm thicknesses, pattern types, and pattern sizes, showing good versatility.
• Based on the finite element model, a large-scale, high-yield, and high-uniformity three-dimensional configurable structure was successfully achieved.
• A series of three-dimensional structured photodetectors have been developed to detect the angle of incident light with an accuracy of 10°, demonstrating their potential for the fabrication of electronic and optoelectronic devices.

Paper address:
https://www.nature.com/articles/s41467-024-47405-2

Release of pre-strained bilayer nanofilm

The research team established a standard strain nanomembrane release model.It consists of a rectangular nanomembrane with width W, length L and in-plane strain gradient, as well as a sacrificial layer and a substrate. This process is designed to ensure the efficient processing and application of nano self-assembled three-dimensional devices, and it also needs to expose more influencing factors that must be considered in finite element modeling.

Figure b shows the release process of the fixed edge pre-strained double-layer nanofilm. Figure c further reveals the wet etching process of the sacrificial layer, where the etchant diffuses from the liquid phase to the solid-liquid interface and reacts chemically with the sacrificial layer at the interface.

Multi-level quasi-static finite element analysis method helps

The research team used the COMSOL Multiphyics software package to establish a multi-field coupled finite element model to simulate the etching process within the sacrificial layer, using a constrained initial fixed bilayer to simulate the unreleased state.

Considering the etchant flow and the boundary motion of the etching system, the research team introduced Fick's law, Navier-Stokes equation and liquid-solid reaction model to establish a finite element model:

Where c is the etchant concentration, u is the etchant flow velocity, D is the diffusion coefficient, k is the reaction rate coefficient, n is the normal vector, ρ is the density, μ is the dynamic viscosity coefficient, v is the normal mesh velocity, and M is the molar mass.

The study discussed the influence of various parameters.It was determined that the relationship between etchant concentration and reaction rate depends on diffusion equilibrium and chemical reaction process.

The entire dynamic release process of the double-layer nanofilm is a series of analysis steps performed in chronological order. In the figure above, b shows the geometric characteristics of the boundary changes obtained from the previous simulation in coordinate form, and the number of nodes for feature division along the thickness direction of the Si/Cr double layer is i = 1, 2, .... Through geometric division, the boundaries at multiple discrete time points are divided into boundary conditions. In the elastic mechanics analysis step, the boundary conditions will be set as a series of constraints in chronological order to achieve the continuous release of the nanofilm, as shown in c in the figure above.

Due to the asymmetric boundary conditions of the double-layer nanomembrane structure fixed at one end, the complex actual release process and large geometric deformation, the structural design using the large perturbation equation of the elastic thin plate cannot accurately reflect the process. The three-dimensional assembly of the nanomembrane released only from the opposite edge can only present a tubular structure.The simulation results of the double-layer nanofilm showed better accuracy and successfully displayed the morphology that varies with width.(As shown in d in the figure above).

The results of quasi-static finite element analysis prove that the etching trajectory of the sacrificial layer plays a vital role in regulating the release of the double-layer nanomembrane. At the same time, in addition to the silicon (Si) / chromium (Cr) rectangular pattern, the research team also used the multi-layer finite element model to design various types of patterns, such as semicircular, triangular and parallelogram patterns. The method also demonstrated the cross-scale compatibility of models from hundreds of nanometers to hundreds of micrometers.

Deep neural networks show their prowess, bringing innovation to 3D angle light detection

The research team selected the Si/Cr double-layer nanofilm system to verify the design model because Cr can introduce a large pre-strain and Si is the most widely used semiconductor material. The experimental process is shown in the figure below.

The multi-level design method based on finite element etching and elastic mechanics modeling achieved the three-dimensional assembly of pre-strained nanofilms, providing new opportunities for forward-looking device design through strain engineering. The research team finally conducted a wide range of incident light angle detection on photodetectors of different shapes.The results show the sensitivity of photoelectric coupling in three-dimensional configurations and can be extended and used to realize the functionalization of three-dimensional assembled nanomembrane electronic devices.As shown in the figure below.

The research team designed a hemispherical omnidirectional incident light controller consisting of a transparent PMMA shell and an optical fiber interface array.The laser can be incident at a specific angle through an interface connected to the corresponding coordinates of the spherical surface, as shown in Figure c above, and the coordinates (θ, φ) are set.

The prepared Si/Cr photodetector is then placed on a platform at the same height as the bottom of the incident light controller, and the controller is calibrated to ensure that the projection coordinates of the (90°, 0°) input laser port on the YZ plane are aligned with the center of the photodetector (as shown in Figures a and b above). Finally, the collected light detection data is imported into the deep neural network for incident angle analysis (as shown in Figure c above).

Specifically, the Si/Cr photodetector has a maximum responsivity of 60mA/W, a response time of 100 ~700 μs, and an external quantum efficiency of 7 – 12 %, which can effectively cope with 520 nm incident light to achieve photodetection, as shown in the figure above.

After measuring the photoresponse of incident light from different coordinates, the research team normalized the photocurrent of each coordinate and classified it by structure type. At the same time, in order to facilitate intuitive comparison of data, the normalized photocurrent of each structure will be projected on the YZ plane through the photocontroller. The research team also established spherical coordinates represented by θ and φ to help identify the position of the incident light on the projected sphere.

The results show that the unreleased Si/Cr planar photodetector (f in the above figure), ring photodetector (g in the above figure), arch photodetector (h in the above figure), helix photodetector (i in the above figure), tapered photodetector (j in the above figure), and tubular photodetector (k in the above figure) show different results for incident angle detection.

Compared with the unreleased Si/Cr planar photodetector, the annular and tubular photodetectors have a wider high photocurrent detection angle, confirming that the three-dimensional microstructure is conducive to the anisotropic detection of incident light. Compared with the annular structure, the tubular photodetector shows better stability in wide-angle detection.

Then, based on the differences in the incident angle detection of the structures in the above experiments,The research team used deep neural networks to establish an angle-sensitive detection model based on photodetectors with different structures.

The research team collected data from 275 incident points for each photodetector in Figures f-k and constructed a three-dimensional spatial incident light photoelectric response database. Using deep neural networks to train longitude and latitude data sets, the accuracy of 95% and 78% was demonstrated respectively (l in the figure above).Based on the above technology, the accuracy of 83% and 71% can be achieved in the detection of longitude and latitude of the incident light direction.The incident light angle resolution reaches ~10°, and its angle prediction weight is significant as can be seen in Figure n above.

The three-dimensional light detector combined with deep neural network and nanofilm assembly can achieve high-precision prediction of the incident light angle.This technology has shown valuable application potential in wearable devices, smart furniture, and smart driving systems.

Young talents are full of vigor and vitality, and microelectronic devices are welcoming new developments

The co-first authors of this paper are doctoral student Zhang Ziyu and Dr. Wu Binmin.Zhang Ziyu is a 2021 doctoral student in the Department of Materials Science at Fudan University. As the first/co-first author, he has published 3 papers and monograph chapters in journals such as Nature Communications and Advanced Materials, participated in the publication of 9 papers and monograph chapters, and applied for 2 Chinese invention patents.

It is worth mentioning that at the time of the release of this research result, Zhang Ziyu was also successfully selected into the first batch of National Natural Science Foundation Young Students Basic Research Project (doctoral students) funding list.

Wu Binmin has also made great achievements. A few months ago, Wu Binmin published a paper titled "One-step rolling fabrication of VO2 tubular bolometers with polarization-sensitive and omnidirectional detection" in "Science Advances" as the first author. The research team discovered that the on-chip tubular nanofilm has excellent thermal insulation and light trapping effects.

This research is not the first collaboration between Zhang Ziyu and Wu Binmin. Earlier this year, the two published a paper titled "Self-rolled-up ultrathin single-crystalline silicon nanomembranes for on-chip tubular polarization photodetectors" in "Advanced Materials" as co-first authors.This research focuses on self-curling single-crystal silicon nanofilm polarization photodetectors.

In addition to the outstanding authors, the progress of two-dimensional nanomaterials is also worthy of attention. Especially in today's world where devices are becoming miniaturized and intelligent, electrical devices that construct complex three-dimensional structures at the nano and micro scales have attracted the attention of major laboratories around the world.

The authoritative journal Nature Communications has collected a number of research results related to this aspect. For example, as early as 2019, Felix Gabler et al. published a research result titled "Magnetic origami creates high performance micro devices", which mentioned an alternative method for assembling high-aspect-ratio nanofilms into microelectronics.The chip manufacturing process can be improved by remotely assisting its assembly through the influence of external magnetic fields.
Paper link:https://www.nature.com/articles/s41467-019-10947-x

The latest collaboration between Zhang Ziyu and Wu Binmin combines deep neural networks and nanofilms to assemble three-dimensional light detectors to achieve accurate analysis of the angle of incident light. By integrating advanced AI, it accelerates the arrival of the More-than-Moore era and promotes the advanced development of optoelectronic applications.