Robotic chip-scale nanofabrication for superior consistency

TL;DR

The paper addresses the variability inherent in academic nanofabrication by introducing a robotic arm system that automates low-volume, high-stakes tasks such as resist development. By combining camera-based chip detection, torque-based height calibration, and a tweezer-based wet-bench workflow, the approach achieves significantly improved repeatability in resist development, demonstrated on Dolan-bridge Josephson Junctions where robotized processing reduces sample-to-sample resistance spread to about $2\%$ versus approximately $7\%$ with human operators. The results show that robotic automation eliminates inter-operator variability, enhancing reproducibility and enabling more reliable exploration of process variations in research settings. The work suggests broad potential for robotic nanofabrication to provide industrial-level consistency in flexible, high-mix research environments and outlines future directions such as LLM-assisted scripting and real-time supervision for robustness.

Abstract

Unlike the rigid, high-volume automation found in industry, academic research requires process flexibility that has historically relied on variable manual operations. This hinders the fabrication of advanced, complex devices. We propose to address this gap by automating these low-volume, high-stakes tasks using a robotic arm to improve process control and consistency. As a proof of concept, we deploy this system for the resist development of Josephson junction devices. A statistical comparison of the process repeatability shows the robotic process achieves a resistance spread across chips close to 2%, a significant improvement over the ~7% spread observed from human operators, validating robotics as a solution to eliminate operator-dependent variability and a path towards industrial-level consistency in a research setting.

Figures (6)

• Figure 1: Robotic Assisted Nanofabrication. Robotic arms equipped with tweezers are the missing link for explorative nanofabrication at sample scale. We foresee four tasks enabled by picking up and handling samples that promise improvement in reliability and safety. (a) Cleaning samples in ultrasonic solvent baths or chemical cleans. (b) Spinning resist using electronic pipettes for volume control, followed by standard spinners and hotplates. (c) Loading prepared substrates on carrier wafers for processing and using resist or oil for bonding. Accurate placement on tool carriers for lithography reduces loading time and increases tool utilization. (d) Developing samples by controlled liquid chemical exposure and final drying step. Explored in this work.
• Figure 2: Robotic Resist Development.(a) After chip detection and hand-eye transformation, the tool-height is calibrated using torque-based feedback. Each chip is then individually picked up, swirled inside a developer solution and a rinse solution for 40 and 10 respectively, before being blow-dried for 10 in a nitrogen flow. Afterwards the chip is returned to the chip-carrier. (b) OpenCV-based chip-edge detection allows free placement of chip-carrier by a human operator. (c) Reliable height detection for chip pick up. Relative torque on joint $5$ of the arm versus vertical step height, starting from an overdriven position (verified by an initial drop in torque of 15 upon lifting the tweezers, see dashed line). The arm raises the tweezers in 0.1 steps until the torque change between steps drops below 0.8 (blue line), inferring that the tips have lifted off the chip box. This point establishes the reference height for grasping. (d) Mecademic M500 Robot on an optical breadboard with access to the required fabrication steps for sample based wet-bench tasks. (e) CAD of the chip showing the array of JJ test devices. In each row, there are $12$ identical devices, and we sweep the JJ width over the $9$ rows. (f) Microscope image of a fabricated JJ test device after probing. Scale bar: 100.
• Figure 3: Robotic vs. human fabrication consistency.(a) Intra-chip consistency of junction resistance. Box plot of the intra-row coefficient of variation (CV) for all device rows, grouped by operator. The red line is the median, the blue box spans the interquartile range (IQR, 25th to 75th percentile) and the whiskers extend to $1.5 \times$ IQR. Individual data points are overlaid as black dots. The robot group exhibits a lower median CV compared to the 'All Humans' group. (b) Inter-chip consistency of junction resistance. Bar chart comparing the CV for the robot (orange) and 'All Humans' (blue) groups for different junction widths. The CV is computed from the set of 9 mean resistances (one from each chip) per operator group for a given junction width.
• Figure Supplementary Figure 1: Chip location during fabrication.(a) Location of the chips on the original $4$-in wafer during e-beam lithography, sorted by operator during developing. HA (B, C) stands for Human A (B, C) and R for Robot. (b) Chips glued to a carrier wafer for the aluminum deposition step. The carrier wafer has a pre-diced grid to facilitate alignment of the chips to the tilting axis during evaporation.
• Figure Supplementary Figure 2: Josephson junction resistance.(a) Josephson junction design, showing what we refer to as JJ width. (b) Mean JJ resistance vs JJ width for the 18 chips, separated into robot (orange) and human (blue) operators. Mean over all the robot (human) chips are marked with an orange (blue) diamond, and the overall mean with a black diamond marker. (c) Mean resistance on each chip normalized by the respective median, and grouped by junction width and operator (black data points). A box plot is overlaid on top of the points showing that the robot group has a smaller resistance spread across chips than the 'All humans' group. The horizontal black line in the box plot is the median.