Operating principle of ACNS for complementary control.
Figure 1a depicts a biological control system based on the complementary signals of SNS/PSNS. The combination of these systems results in complementary stimuli delivered to organs, such as promotion-inhibition and dilation-contraction. Although a biological control system can effectively control the internal environment of the human body through complementary signals, the presence of SNS and PSNS necessitates two independent signal transmission channels, which leads to a high system complexity. In contrast, the proposed ACNS combines the two complementary signals into an integrated signal before it reaches the associated organs, thereby simplifying the transmission system to realize efficient signal transmission, as shown in Fig. 1b.
The operating characteristics of the SIC are of significance in integrating the two complementary signals: Changes in individual signal types can be discerned to reflect each signal in the form of positive and negative amplitudes of the pulses. Figure 1c shows a circuit diagram of the SIC that integrates two complementary signals into voltage pulses. The circuit contains two SBTs connected in series in opposite directions. When the complementary input signals (VG1 and VG2) to be integrated enter the gate of each SBT, an AC supply pulse applied to the drain (VIN) is converted to an asymmetric AC output pulse, the positive and negative amplitudes of which are adjusted according to each input signal. The varying amplitude the two signs of the asymmetric AC can be attributed to the tunable diode behavior of the SBTs. The upper panel of Fig. 1d and Supplementary Fig. 1 show schematic structure and optical microscopy image of two SBTs connected in series, respectively. Each SBT has a graphene/n-type semiconductor/metal heterojunction (Supplementary Fig. 2). Such this structure allows the modulation of the SB, as shown in the lower panel of Fig. 1d. In the forward bias mode, the injection of electrons from the metal into the semiconductor is not affected because the work function change of the metal owing to VG is negligible. In contrast, in the reverse mode, the electron injection can be adjusted based on VG because the SB at the graphene/semiconductor interface can be either raised or lowered by applying negative or positive VG, respectively36–39. The voltage polarity of the SBTs determines the sign of the output current that can be adjusted, as shown in Fig. 1e. In the case of SBT1 (left panel of Fig. 1e), the reverse mode occurs in the negative drain voltage (VD) region. Thus, the positive component of the AC supply pulse yields an output with a constant and high-level positive amplitude, whereas the negative supply component results in VG-dependent negative output pulses. In the case of SBT2 (middle panel of Fig. 1e), the bias condition is reversed: The positive component of the AC supply pulse solely contributes to the positive amplitude of the asymmetric AC pulse, which varies according to VG. When these two SBTs are connected in series (right panel of Fig. 1e), the SBT with the higher resistance determines the amplitude of the output pulse for a given AC supply pulse. As a result, the negative and positive amplitudes of the resulting asymmetric AC output pulse can be modulated only through VG1 and VG2, respectively.
Figure 1f shows the operational diagram of regulation of blood glucose level by the ACNS. First, the glucose and insulin sensors detect each target substance to convert it into an electrical signal. Subsequently, each signal is integrated to voltage pulses including the information of both the glucose and insulin by SIC. Next, the integrated signal enters the AS, which potentiates and depresses the PSC according to the positive and negative amplitudes of the input signal, respectively. When the level of PSC exceeds the set standard, a logical decision is made to suppress the change in the PSC by the ANC. Consequently, a negative feedback is achieved by transmitting the signal to the actuators that inject glucose or insulin to maintain the glucose level in a desirable range.
Electrical properties of single SBT and integration of complementary signals by a complementary input interface.
Figure 2a shows the cross-section and top-view schematic of a single SBT. First, an indium-tin-oxide (ITO) drain electrode and indium-gallium-zinc-oxide (IGZO) channel were sputtered onto a heavily-doped Si/SiO2 substrate. Subsequently, the monolayer graphene was transferred and patterned onto the IGZO channel layer to form the SB between the layers (Supplementary Fig. 3 and 4). Figure 2b and 2c show transfer curves of the SBT under reverse VD ranging from +0.1 V to +1.0 V in 0.1 V increments and forward VD ranging from -0.1 V to -1.0 V in -0.1 V increments, respectively (also see the band diagram and output curve are presented in supplementary Fig. S5 and S6, respectively). In the reverse bias mode, the maximum on-off current ratio of 156 was observed when VD was +0.2 V. However, in the forward bias mode, the on-off current ratio did not exceed 20 under the same VD range (Supplementary Fig. 7). The rectification ratio, which is defined as the ratio of the forward current to the reverse current at VG = 0 V and 40 V as a function of VD is plotted in Fig. 2d and Supplementary Fig. 8. When VG = 0 V, the highest rectification ratio was observed at VD = 0.4 V while the ratio approximately converged to 1 under VG = 40 V. For an SBT that exhibits a high rectification ratio, the contribution of the forward mode SBT to the overall output current of the system in which two SBTs in forward and reverse modes are connected in series is negligible. Because the forward mode current of the adopted SBT does not overwhelm the reverse mode current, it is necessary to design the system to ensure that the difference between the forward and reverse mode currents is maximized. Therefore, VD = 0.4 V, which corresponds to the maximum rectification ratio, was selected as the amplitude of the AC supply pulse of the SIC.
Figure 2e shows the circuit diagram of the complementary input interface, which consists of a glucose sensor, an insulin sensor, and a SIC that incorporates the electric signals from the two sensors. Transistor-type sensors are fabricated for detecting the glucose/insulin level. IGZO was used as a benchmark sensing channel despite its lack of selectivity because the current level can be varied by several orders according to the amounts of glucose or insulin solutions (Supplementary Fig. 9 and 10). The high on-off current ratio of transistor-type sensors allows three glucose/insulin levels to be distinguished. The amounts of glucose and insulin detected by each sensor are first converted to the corresponding Isensor, as shown in the left panel of Fig. 2f. States “0”, “1”, and “2” denote the number of glucose or insulin doses dropped on each sensor, respectively. Next, Isensor was converted to a voltage signal via a voltage divider followed by voltage amplification. By selecting the proper reference resistance and voltage gain of the amplifier, it is possible to adjust Isensor to the intended voltage level. In the considered case, states “0”, “1”, and “2” correspond to approximately 0 V, 20 V, and 40 V, respectively, as shown in the right panel of Fig. 2f. Finally, the amplified voltage was applied to the gate of each SBT connected in series in opposite directions while the AC supply voltage, with an amplitude of 0.4 V and frequency of 2 Hz, was applied to the drain of the SBT. During the positive amplitude phase, the SBTs connected to the glucose and insulin sensors were in the reverse and forward modes, respectively. Therefore, the amplitude of the positive composition of the integrated output voltage (VIO) pulse was determined by VG1 of the SBT connected to the glucose sensor (i.e., the state of glucose). Likewise, the negative composition of the VIO pulse was regulated by the state of insulin. As a proof of concept, the VIO pulse according to the states of glucose and insulin were measured, as shown in Fig. 2g. As the glucose and insulin state increased, the amplitude of the positive and negative components of the VIO pulse increased, respectively, and the state of each substance independently contributed to the amplitude of the VIO pulse.
Synaptic behavior of AS under integrated pulses.
In order to regulate the levels of glucose and insulin within a desirable range, it is necessary to first understand the synaptic behavior of the AS induced by the VIO pulse input. Figure 3a depicts the structural schematic of the AS and VIO entering the AS. For the fabrication of the AS, a poly(3-hexylthiophene) (P3HT) and an ion-gel were patterned onto the substrate with Au electrode patterns (Supplementary Fig. 11). Detailed working mechanism of the P3HT/ion-gel AS is as follows: Negative VIO pulses drive the penetration of [PF6]- anions in the ion-gel into the free volume of the P3HT channel layer and facilitate the doping of the P3HT channel with the anions to increase the PSC. Although the penetrated anions gradually diffuse out from the channel in the absence of the negative VIO pulses, the residual anions in the P3HT prevent a drastic reduction in the PSC, resulting in retentive current characteristics. Furthermore, the penetrated [PF6]- ions are extracted by positive VIO pulses, and current retention occurs during the absence of VIO. Such retentive characteristics of the P3HT/ion-gel layer results in the synaptic behavior23. Figure 3b shows the normalized EPSC and IPSC of the P3HT/ion-gel AS. VIO pulses with magnitude from +1 V to +3 V and a width of 40 ms were applied to the ion-gel, and clear EPSC and IPSC responses are observed at a VD of 0.5 V. Moreover, the AS exhibited reliable long-term potentiation (Fig. 3c) as well as long-term depression behaviors (Fig. 3d). Long-term plasticity due to the retentive ion movement in the P3HT channel was observed up to 50 consecutive pulses, and the changes in the PSC became larger as the pulse magnitude increased.
Based on the fundamental synaptic performance discussed in the previous sections, the PSC change of the P3HT/ion-gel AS by VIO pulses was clarified, as shown in Fig. 3e. Prior to applying VIO pulses having both positive and negative peak components, the current potentiation by negative pulses with a magnitude of -3 V (V0,3) was investigated. As shown in the first panel (black line), gradual PSC increase was observed. Next, the synaptic device subjected to 30 V0,3 pulses in advance was exposed to continuous VIO pulses, with the magnitude of the positive peak component set as +1 V (red line), +2 V (blue line), and +3 V (green line), and the magnitude of the negative peak component set as -3 V (denoted as V1,3, V2,3, and V3,3, respectively). Compared to V0,3, V1,3 pulses mitigated the increase in PSC, as indicated by the gray shaded region. In the case of V2,3, the PSC decreased from the initial state but maintained a certain current level even when the number of pulses increased. The application of V3,3 further decreased the PSC level, reaching nearly zero after nine pulses. Additional experimental results regarding the PSC changes under positive peak components of 0.5 V, 1.5 V, and 2.5 V along with 0 V, 1 V, 2 V, and 3 V are shown in Supplementary Fig. 12. Supplementary Fig. 13 shows the derivative of PSC with respect to the number of pulses. When the number of pulses is sufficiently large and the positive peak component is less than 1 V, the derivative was positive; above 2 V, the value was close to zero. Figure 3f indicates the level of normalized PSC according to the magnitude of the positive peak component. As the depression voltage increased from 0 V to 3 V, the PSC after 30 pulses decreased from 2.81 to 9.4 × 10-3. These results indicate that the PSC behavior can be controlled by modulating the positive and negative amplitudes of the VIO pulses.
To construct a glucose level-regulating system, a Schmitt-trigger-based ANC that provides feedback regarding the glucose and insulin levels was designed. The ANC consists of four electrical parts: i) a resistor-capacitor delay and a non-inverting amplifier that reduces the intensity of the spike component of the synaptic output voltage (VSO, ANC-entering voltage converted from PSC) and amplifies the mitigated VSO (i.e., VRC), respectively; ii) a Schmitt trigger that outputs the voltages (VSCH) for the VRC according to two set thresholds of VUT and VLT; iii) a differentiator that converts the square wave of VSCH into a pulse wave (VDIFF); and iv) a PNP and NPN bipolar-junction transistor (BJT) that selectively transfers negative or positive pulses (ANC output voltage, VNO). These VNO signals induce an injection of glucose and insulin to adjust the glucose level.
Schmitt-trigger-based ANC for glucose-regulating systems.
Since the number and signs of the output pulses depend on the number of Schmitt triggers and the type of BJT, a variety of glucose-level-regulating systems can be implemented by varying these two electrical parameters. Figure 4a shows the simplified circuit diagram of a double-negative spike generation ANC having two Schmitt triggers with different reference voltages (VREF) and a PNP BJT. Utilization of two Schmitt triggers connected in parallel allows three distinct levels of VSCH to be implemented, thereby yielding double VDIFF spikes. Moreover, the PNP BJT selectively passes only negative pulses. The concept of the double-negative spike generation mode glucose-level-regulating system is demonstrated in Fig. 4b, which shows the output voltages obtained after traversing each electrical component. As the input, triangular pulses (VAmp = 10 mV, f = 0.2 Hz) were applied with a supply voltage of ± 10 V and VREF of 4.5 V and 6.5 V. The two pairs of thresholds set according to the Schmitt triggers were VUT = 6.5 V, VLT = 2.5 V and VUT = 8.5 V, VLT = 4.5 V for the double VDIFF spike. VSCH exhibits an upshift and downshift when VRC coincides with VUT and VLT, yielding a negative and positive pulse, respectively; this phenomenon is followed by the filtering of the positive pulses by the PNP BJT. In this manner, double-negative VNO spikes capable of inducing a steeper increase in the glucose level could be generated. In addition to the double-negative spike generation ANC, a complementary spike generation mode glucose-level-regulating system was designed by replacing the pair of Schmitt-triggers with a single one and adding an NPN BJT. The operation principle of this system is identical to that of the abovementioned system, except for the VDIFF spike generation and pulse filtering being performed by the BJT, as shown in the simplified circuit diagram of the complementary spike generation ANC (Fig. 4c and 4d). By using one Schmitt-trigger (VUT = 7.5 V, VLT = 3.5 V) and both types of BJT, a single VDIFF spike in both positive and negative can be obtained. The detailed circuit diagrams of the ANCs are shown in the supporting information (Supplementary Fig. 14 and 15). The advantage of this system is that both glucose and insulin injection can be facilitated in a manner that prevents a radical change in the glucose level.
Glucose-level-regulating systems controlled by integrated signals.
Finally, glucose-level-regulating negative feedback loops were constructed by incorporating the functions of the sensor, SIC, AS, and ANC. Two types of glucose-level-regulating feedback loops, characterized by the abovementioned ANC, were demonstrated: a double insulin injection feedback loop, and an insulin-glucose injection feedback loop. Figure 5a shows the logical algorithm of the double insulin injection feedback loop. Unless VRC exceeds VUT1, no feedback regarding the glucose level is required. When VRC reaches the first upper threshold (VUT1), which corresponds to an abnormal increase in the glucose level, the ANC outputs a negative pulse to promote insulin injection. After the first injection, VRC can either remain in the range below VUT1 or further increase to reach the second threshold (VUT2). The former case indicates that the amount of injected insulin is sufficient to stabilize the glucose level. In the latter case, additional insulin is injected to induce a stronger suppression of glucose level increase, leading to glucose level stabilization. Figure 5b exemplifies the former case in which VRC with an initial glucose state of 1 changes according to the generated VNO. When the increasing VRC exceeded VUT1, one dose of insulin was injected, and no additional injection is performed. The latter case with an initial glucose state of 2 is shown in Fig. 5c. VNO generated by the two thresholds effectively suppressed the rapid increase in VRC. The logical algorithm of the insulin-glucose injection feedback loop is shown in Fig. 5d. When VRC approaches VUT or VLT and crosses each threshold, the injection of insulin and glucose, respectively, is implemented to suppress the corresponding change. The regulation of VRC by VUT and VLT is demonstrated in Fig. 5e, successfully confining the level of VRC to within the intended voltage range.