Source-linked AI summary
Large-Scale Integrated Flexible Tactile Sensor Array for Sensitive Smart Robotic Touch
Zhenxuan Zhao, Jianshi Tang, Jian Yuan, Yijun Li, Yuan Dai, Jian Yao, Qingtian Zhang, Sanchuan Ding, Tingyu Li, Ruirui Zhang, Yu Zheng, Zhengyou Zhang, Song Qiu, Qingwen Li, Bin Gao, Ning Deng, He Qian, Fei Xing, Zheng You, Huaqiang Wu
TL;DR
Robotic touch calls for large-scale integrated pressure arrays with high spatial resolution to map contact. The authors built a flexible 64×64 tactile array that reached 0.9 mm spatial resolution and achieved 98.8% handwritten-digit testing accuracy in hardware.
Problem
Robotic touch requires large-scale integrated pressure arrays with high spatial resolution to detect contact location, shape, size, and pressure distribution.
Source: RESULTS AND DISCUSSION · p. 14Method
The authors integrated a high-performance piezoresistive film with a 64×64 carbon-nanotube thin-film-transistor active matrix to build a flexible tactile sensor array.
Source: RESULTS AND DISCUSSION · p. 14Results
0.9 mm spatial resolution (28.2 ppi) was achieved, and handwritten-digit testing accuracy reached 98.8% in hardware.
Source: RESULTS AND DISCUSSION · pp. 15, 21Takeaways & Limitations
The work demonstrates a 4096-pixel, high-resolution integrated flexible pressure-sensor array for smart robotic touch.
Source: CONCLUSION · p. 22
Abstract
from arXiv · showhide
In the long pursuit of smart robotics, it has been envisioned to empower robots with human-like senses, especially vision and touch. While tremendous progress has been made in image sensors and computer vision over the past decades, the tactile sense abilities are lagging behind due to the lack of large-scale flexible tactile sensor array with high sensitivity, high spatial resolution, and fast response. In this work, we have demonstrated a 64x64 flexible tactile sensor array with a record-high spatial resolution of 0.9 mm (equivalently 28.2 pixels per inch), by integrating a high-performance piezoresistive film (PRF) with a large-area active matrix of carbon nanotube thin-film transistors. PRF with self-formed microstructures exhibited high pressure-sensitivity of ~385 kPa-1 for MWCNTs concentration of 6%, while the 14% one exhibited fast response time of ~3 ms, good linearity, broad detection range beyond 1400 kPa, and excellent cyclability over 3000 cycles. Using this fully integrated tactile sensor array, the footprint maps of an artificial honeybee were clearly identified. Furthermore, we hardware-implemented a smart tactile system by integrating the PRF-based sensor array with a memristor-based computing-in-memory chip to record and recognize handwritten digits and Chinese calligraphy, achieving high classification accuracies of 98.8% and 97.3% in hardware, respectively. The integration of sensor networks with deep learning hardware may enable edge or near-sensor computing with significantly reduced power consumption and latency. Our work could pave the road to building large-scale intelligent sensor networks for next-generation smart robotics.
RESULTS AND DISCUSSION
The work demonstrates a flexible 64×64 PRF-based tactile sensor array and a smart tactile system that combines the array with memristor computing-in-memory hardware to recognize handwritten patterns. Results include high-resolution pressure mapping, strong pressure-sensing performance, and high recognition accuracy for handwritten digits and Chinese characters.
High-performance pressure sensor with PRF
- The PRF has a rough, self-formed microstructured upper region over a flatter MWCNT/TPU composite region; compression increases electrode contact and current pathways at low pressures, then compresses the lower composite at higher pressures.The top and bottom surfaces had RMS roughnesses of 8.4 μm and 0.2 μm, respectively.Source: RESULTS AND DISCUSSION · pp. 8–10
- At 0~15 kPa, PRF sensitivity was 385 kPa-1, 158 kPa-1, and 44.4 kPa-1 for 6%, 9%, and 14% MWCNTs; 14% PRF showed ~10 kPa-1 across 15 ~ 1400 kPa.The 14% PRF also had a broad pressure detection range of 1400 kPa.Source: Figure 3 · p. 11, RESULTS AND DISCUSSION · p. 12
- The 14% PRF responded in about 3~5 ms and switched reliably through tests up to 3000 cycles; 14% and 6% films detected robot-dog trotting and post-exercise pulse changes, respectively.The dog-trotting test reached about 1000 kPa; pulse monitoring detected a change from ~70 beats/min before exercise to ~101 beats/min afterward.Source: RESULTS AND DISCUSSION · pp. 11, 13, 14
High spatial resolution flexible pressure sensor array
- The flexible active matrix reached 64×64 pixels at 0.9 mm (28.2 ppi), with 4083 of 4096 TFT pixels functional (99.7%) and an integrated-array response time of ~4 ms.An individual integrated pixel showed over 1000× output current dynamic range from 0 to ~118 kPa, with extracted pressure sensitivity of 42.5~12.9 kPa-1.Source: RESULTS AND DISCUSSION · pp. 15, 17, Figure 4 · p. 16
- The array correctly identified all six feet of a 6.7-gram artificial honeybee and mapped differing pressures across eight pixels under its hindfoot, whose stripe pattern matched the foot’s shape and size.The hindfoot pixel current varied from 1.83~0.084 µA, reflecting increased landing force from top to bottom.Source: RESULTS AND DISCUSSION · pp. 17–18
Smart tactile system for sensing and recognition
- Handwritten-digit testing accuracy reached 98.8%, while Chinese-character recognition reached 97.3±1.0% with 90% handwritten samples in training, near the 97.6±1.2% software baseline.The prototype integrated the 64×64 sensor array with a memristor-based computing-in-memory chip; the Chinese-character results were averaged over five experiments.Source: Figure 5 · p. 19, RESULTS AND DISCUSSION · pp. 20–21
CONCLUSION
The study demonstrates a large-scale integrated flexible tactile array and a memristor-based smart tactile system that senses and recognizes inputs. The array combines high sensitivity, broad pressure detection, fast response, and fine spatial resolution; the hardware system recognizes handwritten digits and Chinese characters with high accuracy.
- 98.8% and 97.3% classification accuracies were achieved in hardware experiments for handwritten digits and Chinese character recognition, respectively.The prototype integrated the 64×64 pressure sensor array with a memristor-based computing-in-memory chip.Source: CONCLUSION · p. 22
- The 4-inch, 64×64 CNT TFT active matrix achieved 99.7% yield, while the integrated array provided 4096 pixels at 0.9 mm spatial resolution.The PRF exhibited sensitivity up to 385 kPa-1, a 0~1400 kPa detection range, 3~5 ms response speed, and cyclability over 3000 cycles.Source: CONCLUSION · p. 22
- Self-formed microstructures on the PRF surface were suggested to underlie its pressure-sensing performance, and the array demonstrated honeybee footprint-map and human wrist-pulse detection.The authors highlight the array's potential for edge or near-sensor computing in future robotics.Source: CONCLUSION · p. 22
EXPERIMENTAL SECTION
The experimental section describes fabrication of the MWCNT/TPU piezoresistive film (PRF), its integration with a CNT TFT active matrix, and fabrication of a memristor-based computing-in-memory (CIM) chip.
Fabrication and characterization of MWCNTs/TPU composite as PRF
- MWCNT/TPU PRFs with 6%, 9%, and 14% MWCNT weight ratios were prepared by casting a precursor onto a substrate and baking at 90 °C for 4 hours.TPU was dispersed in DMF at a 1:5 weight ratio before adding the MWCNT dispersion; the section notes that DMF critically affects PRF preparation and pressure-sensing performance.Source: EXPERIMENTAL SECTION · pp. 22–23
Fabrication of the CNT TFT-based active matrix
- A 64-line-gate and 64-line-drain CNT TFT active matrix was fabricated on polyimide, covered with 14% MWCNTs/TPU PRF, and delaminated as a flexible sensor array.The TFT channels were 8 μm long and 100 μm wide; the PRF precursor was cast over the full active matrix.Source: EXPERIMENTAL SECTION · pp. 23–24
Fabrication of memristor-based CIM chip
- A commercial-foundry 130 nm CMOS process fabricated the CIM chip with a 784×100×10 MLP and approximately 160K monolithically integrated TiN/HfOx/TaOy/TiN memristors.The chip included ADCs, DACs, drivers, and buffers; each synaptic weight used a differential pair of 1T1R cells.Source: EXPERIMENTAL SECTION · p. 24