Mega 2560

A programmable 48‐channel voltage pulse generator was constructed using three 16‐channel evaluation boards coupled to a myRIO device (Figure S6, Supporting Information). A custom LabVIEW program was used to define the parameters of the voltage pulse sequence generated by the device.
A TTP229 (TonTouch) 16‐key capacitive touch keypad from TonTek was interfaced with an Arduino Mega 2560 microcontroller (ATmega 2560) board, shown in Figure S4, Supporting Information. The Arduino Mega 2560 was programmed to take inputs from the TTP229 touchpad and deliver output pulse trains (5 V) to the synapse array through resistive voltage dividers (82kΩ/10kΩ).
A µA741 operational amplifier from Texas Instruments was configurated as a transimpedance amplifier to bridge the classifier circuit (reading at 50 mV) and the OECN artificial neuron (firing at 0.3–0.5 V), shown in Figure S7, Supporting Information.

Figure 2 shows the overall system integration and control of the hand rehabilitation system, combining the exoskeleton-assisted hand motion and the fingertip cutaneous haptic stimulation. The motor stroke sequence is embedded in an Arduino Mega 2560. When the computer sends a start command, the Arduino Mega 2560 starts to send the control signals to the linear motors in the hand exoskeleton. The motor stroke feedback signals are sent to an analog input/output module JY-DAM10AIAO. The target haptic force is calculated, according to the selected mode and the feedback motor stroke information and transfers to the analog input/output module JY-DAM10AIAO to control the air pressure inside the fingertip cutaneous haptic stimulation actuators via the pressure regulators SMC ITV0010. Pressurized air is provided by an air compressor U-STAR601 as reported before. The feedback signals from the pressure regulators are monitored by the JY-DAM10AIAO device.

For print platform motion, a Nema 57 stepper motor supplied by a 12-V power bank was used to drive vertical build platform translation along a 30.5 cm Stroke Linear Motion router (VXB Ballbearings, Anaheim, CA, USA). The UV light engine used was a 3DLP9000 (Digital Light Innovations, TX, USA) with a 4 million pixel 2560 × 1600 digital micromirror device (DMD), configured with a 385-nm light-emitting diode (LED) and a 30-μm field-of-view projection lens, with a total projection area of 76.8 mm by 48 mm. The light engine is a combination of a DMD chip set (DLP9000, Texas Instrument, TX) along with a projection lens; the intrinsic specification of the DMD chipset is 385-nm UV wavelength, 2560 × 1600 DMD array, 7.6-μm by 7.6-μm pixel size, and build area of 19.5 mm by 12.2 mm; the projection lens diverges the UV projection to a 2560 × 1600 array of 30-μm by 30-μm pixels to a build an area of 76.8 mm by 48.0 mm at a working distance of 126.5 mm. The printer was coordinated with an Arduino MEGA 2560 microcontroller and RAMPS 1.4 shield running open-source Marlin firmware. Custom software, written in C++ and implemented in the Qt Integrated Development Environment to provide a graphical user interface, allowed for tailoring of UV light intensity, UV exposure time, stage speed and acceleration, layer thickness, and delay time after layers, within and between prints.

The output voltage of each element of the tactile sensor array was read by the microcontroller board (Arduino version mega 2560). The reading was performed for each row per time by assigning the logic “1” for the row being read. On the other hand, the logic “0” was set for other rows. Oi=1, O1, …, Oi−1, Oi+1, …, On=0 . Then, the analog port received voltage signals (for the entire row) from the tactile sensor. The logic output was reassigned by shifting “1” to the next row, while other rows were set as “1” to obtain the next scan. This process was repeated until the entire array was scanned. Moreover, logic “0” in the multiplex circuit is grounded and sets the threshold values to cut off the noise for avoiding the crosstalk effect [35 (link),60 (link)]. The data were sent to a computer for data processing, such as obtaining tactile images, training, and testing for image recognition. The operating system is illustrated in Figure 13a, and the image acquisition GUI is illustrated in Figure 13b.

Behavioral boxes consisted of a metal cage (mice, Island Motion, Tappan, NY, USA) or a plastic bucket (rats, IKEA, Alfragide, Portugal) containing one speaker (Cover Industrial Co., Guangdong, China) and three nose ports (Island Motion). Each nose port contained one infra-red beam/sensor pair for detecting nosepoking and one visible LED. The choice ports contained, in addition, a water tube connected to a solenoid valve for reward delivery. Valves were calibrated to deliver 25 or 5 μl of water per reward event (rats and mouse, respectively).
Except for the video camera, all sensors and effectors in the behavioral box were read and controlled by an Arduino Mega 2560 microprocessor (additional information and free software available at http://www.arduino.cc/) via a custom circuit board. The microprocessor implemented the behavioral task, and, through a serial communication port, outputted data to a desktop computer running custom software based on Python's pySerial module (freely available at http://pyserial.sourceforge.net/).

The onboard electronics of the control system consisted of a microcontroller (Arduino Mega 2560) connected to a custom printed circuit board (PCB), a diaphragm pump controller PCB (DFRobot), two diaphragm pumps, three digital pressure sensors (one for high and two for low pressure supply) and 13 three-way normally open solenoid valves. All onboard electronics were powered by a single 12 V power supply. 12 V were used to supply power to the diaphragm pumps, solenoid valves and high-pressure sensor. For the microcontroller and low-pressure sensors, voltage was stepped down with a 12 V to 5 V switching regulator (Recom Power) integrated into the custom PCB. The design of the PCB can be found in the following GitHub repository: https://github.com/RevzinLab/Control-System-for-Microfluidics (accessed on 30 October 2022).

The load cell CM−0.5 kN by X-Senors GmbH, Germany, measures the force change relative to an initial force as a consequence of the volume change during insertion/extraction of the charge transfer species in the active materials. The nominal force of the load cell is $F_{\mathrm{nom}}=$ 0.5 kN within a tolerance of the output signal $s\le \pm 0.2$ % Full Scale (F.S.), a linearity l and hysteresis h of $l=h\le \pm 0.2$ % F.S. A 24-bit analog-digital converter HX711 by Avia Semiconductors converts the analog voltage signal of the load cell, while an Arduino Mega 2560 calculates and stores the pressure $p_{\mathrm{cell}}$ values. In order to calibrate the load cell, we use precision calibration weights. Figure 2 shows the calibration result of the load cell with calibration weights. One can see that the linear calibration line is confirmed by measuring calibration weights. The obtained calibration factor of the calibration line is used by the Arduino software for the calculation of the weight. The resolution, as well as the accuracy of the load cell equates to the measurement precision of the pressure monitoring cell.