Development of neural networks chip generating driving waveform for electrostatic motor

The authors are studying hardware neural networks (HNN) to control the locomotion of the microrobot. The neural networks chip is the integrated circuit chip of the HNN. We proposed the electrostatic motor that is the new actuator of the microrobot in our previous research. The electrostatic motor used the waveform generator to generate the driving waveform. In this paper, the authors will propose the driving circuit using neural networks chip. The cell body model is the basic component of the neural networks chip that outputs 3 MHz frequency of electrical oscillated pulse waveform. Therefore, large capacitors need to connect outside of the neural networks chip to generate the low-frequency driving waveform. The proposal neural networks chip generates a long delay without using large capacitors. In addition, the neural networks chip generated a two-phase anti-phase synchronized waveform by incorporating a mechanism for adjusting synaptic weight. As a result, the proposal neural networks chip can generate the electrostatic motor’s driving waveform with variable frequency. The frequency of the driving waveform could vary from 40 to 126 Hz.


Introduction
Insects have an excellent function. Although insects are small bodies, they perform external recognition through vision and touch. In addition, insects respond to the external environment with excellent control. Insects have a brain, muscles, sensory organs, and energy sources in small bodies.
If the autonomous robots need to be small as insects, each component has to miniaturize. The microrobots are expected to search for a small place where people cannot enter. The centimeter-sized robot "HAMR" developed by Harvard University is equipped with a control circuit and a power supply to achieve independent walking [1]. In addition, the millimeter-sized robot developed by the University of Maryland has successfully walked using an external magnetic field [2]. However, miniaturization of a power supply, sensors, a control circuit, and actuators are a difficult subject [3].
This work was presented in part at the 25th International Symposium on Artificial Life and Robotics (Beppu, Oita, January 22-24, 2020). Programming control by microcontrollers is the dominant system for robot control. On the other hand, insects have excellent sensory information processing and body control, enabling them to adapt to various environments despite their small size [4]. Therefore, research on applying the control method of biological neural networks to robots is conducted [5]. Information processing methods that mimic living organisms have the potential to create simple and compact systems.
The authors are studying hardware neural networks (HNN) [6][7][8][9][10]. The neural networks chip and the body parts of the microrobot were both made by a silicon wafer. The neural networks chip can integrate directly into the body parts of the microrobot. The development of neural networks chip has an advantage because microcontrollers need to mount on a circuit board.
Previously, our constructed microrobot system succeeded to perform walking using an external power supply [7]. The neural networks chip generates the gait pattern of the microrobot. The neural networks chip is the integrated circuit chip of the HNN. A shape memory alloy (SMA) actuator is used for driving the legs of the microrobot. SMA actuator has a large generating force and is simple and easy to miniaturize. However, the power consumption was high. Therefore, we developed an electrostatic motor as a new actuator [11]. Replacing the SMA actuator with the electrostatic motor can achieve a low power consumption of 60 V drive. The microrobot can also drive by solar cells. A variable frequency square waveform is required to drive the electrostatic motor. A waveform generator has been used to generate the driving waveform of the electrostatic motor. Therefore, we proposed HNN to generate the driving waveform of the electrostatic motor to miniaturize the microrobot system [12]. In the simulation result, the HNN generated the driving waveform of the electrostatic motor. In addition, we fabricated the neural networks chip. The neural networks chip successfully generated pulses with a variable frequency (around 50-100 Hz). However, the neural networks chip could not generate a twophase waveform with anti-phase synchronization [13].
In this paper, the authors will propose the neural networks chip that introduced a variable inhibitory-synaptic model. The two-phase waveforms can synchronize as anti-phase by adjusting the synaptic weight. In addition, we will discuss the measuring result of the neural networks chip. Figure 1 shows our previously proposed microrobot system [7]. The external dimensions are 4.6 mm × 9.0 mm × 6.4 mm.

Microrobot
Each part of the robot was manufactured using micro-electro-mechanical systems technology. The SMA actuator drives the microrobot. The neural networks chip mounted on the robot can generate a gait pattern of the microrobot. The neural networks chip realizes miniaturization and weight reduction of the microrobot. Figure 2 shows our previously proposed electrostatic motor [8]. The size of the electrostatic motor was 2.2 mm × 2.5 mm. The electrostatic motor consists of two pairs of electrostatic actuators, a central shuttle, arms, sub-springs, a main-spring, and three-electrode pads, V D1 , V D2 , GND. The arms transmit the force of the electrostatic actuators. The electrostatic motor produces linear motion of the shuttle by energizing the electrodes. Sub-springs and main-spring return the shuttle to the primary position. The electrostatic motor outputs more than 1.3 mN that is suitable to actuate the microrobot's legs.   Figure 4 shows the driving waveform of the electrostatic motor (V D1 , V D2 ). The driving waveform has a pulse width of 7.5 ms, a pulse period of 10 ms, and an amplitude of 60 V. The driving waveform is generated by switching the transistor using a waveform generator. In Fig. 4, V WG1 and V WG2 are waveforms generated by the waveform generator. The authors will replace the waveform generator with the neural networks chip in this paper.

Previous neural networks chip
Previously, we have shown that the neural networks chip can generate the drive waveform of the electrostatic motor shown in Fig. 4. using HSPICE simulations [12]. However, the fabricated neural networks chip's measured results could not perform the two-phase anti-phase synchronized waveform. The reason is the synaptic model that synchronizes the two anti-phase outputs did not work correctly due to parasitic capacitance [13]. Therefore, in this paper, we fabricated a neural networks chip with an additional mechanism to change the synaptic model's weight. The following sections provide details.

Mechanism of the hardware neural networks
If the self-oscillating cell body model (Self-OSC 1) is not connected with the other cell body model, the Self-OSC 1 oscillates at 3.0 MHz. The separately-excited oscillation cell body model S 11 and S 12 are the delay circuit. The delay mechanism is as follows.
1. Self-OSC 1 oscillates a pulse. 2. The pulse of Self-OSC 1 excites the S 11 . 3. S 11 oscillates a pulse. 4. The pulse of S 11 inhibits the Self-OSC 1. It also excites the S 12 . 5. S 12 oscillates a pulse. 6. The pulse of S 12 inhibits the Self-OSC 1.
As a result, Self-OSC 1 could not oscillate a pulse during the inhabitation from S 11 and S 12 . Figure 7 shows an example of a generated waveform of neural networks chip. Figure 7 shows that the mechanism worked, the pulse period increased and the pulse width increased. In addition, S 11 and S 21 inhibit Self-OSC 1 and Self-OSC 2, respectively. Thus, the output of HNN V NN1 and V NN2 will be an anti-phase waveform.  Figure 8 shows the photograph of the fabricated neural networks. A red dotted line indicates the constructed HNN, where the other part is the test element. An IC chip's size is 2.5 × 2.5 mm. 14 electrode pads used as V A , V AS , V DD , V int , V W , V NN1 (output 1), V NN2 (output 2), and GND for the created network part.  Fig. 9., the neural networks chip output a two-phase waveform with anti-phase synchronization. Figure 10 shows an output-frequency characteristic of neural networks chip by varying V int . The plots show the example measurement points. The frequency can be varied linearly from 40 to 126 Hz. In Fig. 10, the voltage of the circuit is as follows. V A = 3.43 V, V AS = 0.80 V, V DD = 3.36 V, V int = 1.90 V. V int is a voltage that determines the degree of delay in the synaptic model with a pulse delay function. Each plot in Fig. 10 is data when antiphase-synchronized waveforms are generated by adjusting the voltage V W of the variable inhibitory-synaptic model. Table 1 shows the characteristics of V int and V W in Fig. 10. A varying only voltage of V int does not output an anti-phase synchronized waveform. By adjusting V W as shown in Table 1, the two pulse outputs synchronize with anti-phase. The accuracy required for the voltage V W is ± 10 mV.

Measurement result
As a result of the neural networks chip measurement, the fabricated neural networks chip generated a two-phase waveform with anti-phase synchronization. The output frequency can be varied from 40 to 126 Hz by changing V int . When the frequency is varied, a two-phase waveform with anti-phase synchronization can be generated by adjusting the voltage V W .

Conclusion
In this paper, the authors developed the neural networks chip that incorporates a mechanism to adjust the synaptic weight. As a result, the proposal neural networks chip generated the electrostatic motor's driving waveform with variable frequency. The frequency of the driving waveform could vary from 40 to 126 Hz.
In the future, we will experiment with driving the electrostatic motor using the neural networks chip. In addition, we will propose the microrobot system using the electrostatic motor.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.