Supplementary Materials for

Transcription

1 advances.sciencemag.org/cgi/content/full/2/6/e /dc1 Supplementary Materials for Organic core-sheath nanowire artificial synapses with femtojoule energy consumption Wentao Xu, Sung-Yong Min, Hyunsang Hwang, Tae-Woo Lee Published 17 June 2016, Sci. Adv. 2, e (2016) DOI: /sciadv The PDF file includes: fig. S1. Schematic of the scalable fabrication of the ion gel gated ONW ST arrays. fig. S2. Schematic of electrical characterization. fig. S3. Core-sheath structure of the ONWs. fig. S4. Schematic of the ion migration in the ion gel and charge carriers in ONW for short-term and long-term plasticity. fig. S5. Control over inter-onw spacing. fig. S6. Control over ONW diameter. fig. S7. SEM image of aligned N2200/PVK (7:3) polymer semiconductor NWs. fig. S8. Short-term synaptic enhancement. fig. S9. Current-retention behavior fits well with the curve of forgetting. fig. S10. Further decreased IPSC was observed with an increased number of presynaptic spikes. fig. S11. Spike timing dependent plasticity. fig. S12. SEM images for the nanogap formation. fig. S13. Configuration of a typical ion gel gated ONW ST. fig. S14. Spike rate dependent plasticity of a short-channel ST. fig. S15. Presynaptic spike voltage dependent EPSC behavior of the ONW ST. fig. S16. Presynaptic spike duration dependent EPSC behavior of the ONW ST. fig. S17. Postsynaptic current in ONW ST triggered by 8-ms pulses. fig. S18. Data selection for SNR calculation. fig. S19. Transistor characteristics of the electronic devices. References (37 43)

2 A. Schematic of the fabrication of ONW synaptic transistor fig. S1. Schematic of the scalable fabrication of the ion gel gated ONW ST arrays. (A) Au/Ti electrodes were patterned on a 300-nm-SiO2 coated highly-doped silicon wafer. (B) Highly aligned ONWs were printed on the pre-deposited electrodes using an e-nw printer. (C) Ion gel was drop-cast onto the channel regions of the devices. B. Schematic of electrical characterization The ONW STs are three-terminal devices. The measurement of the device is demonstrated as follows (fig. S2). Input voltage pulses that emulate presynaptic spikes from a pre-neuron are applied to the metal probe working as a gate electrode. The input pulses cause ions to migrate in the ion gel (main text) to cause change in the source-drain current flowing through the semiconducting ONW. The output current signal that emulates post-synaptic current flowing to a post-neuron in a biological synapse is measured by recording the drain current. fig. S2. Schematic of electrical characterization. During measurement, the ion gel gate is connected to the pre-neuron, the drain electrode is connected to the post-neuron, and the source electrode is grounded.

3 C. Core-sheath structures of ONWs To achieve high resolution in elemental analysis, a relatively thick ONW with a diameter of 300 nm (fig. S3A) was used for EELS analysis (fig. S3, B and C). A P3HT/PEO core-sheath structure was identified using EELS spectroscopy. fig. S3. Core-sheath structure of the ONWs. (A) TEM image of a typical ONW (diameter ~300 nm) that has a core-sheath structure. (B) Elemental mapping by EELS (green: carbon, red: oxygen, yellow: sulfur). (C) A combination of EELS mapping of oxygen (red) and sulfur (yellow) that confirms the core-sheath structure with P3HT inner core surrounded by PEO sheath.

4 D. Schematic of the working mechanism of ONW ST for short-term plasticity and longterm plasticity fig. S4. Schematic of the ion migration in the ion gel and charge carriers in ONW for shortterm (A-C) and long-term plasticity (D-F). (A) Before a presynaptic spike, cations and anions are distributed randomly in the ion gel. (B) During the application of a negative presynaptic spike, anions accumulate near ONW and charge carrier density is enhanced in the ONW. (C) After the presynaptic spike, the accumulated anions gradually drift back to equilibrium positions and the charge carriers in the ONW decay correspondingly. (D) Before a presynaptic spike, cations and anions are distributed randomly in the ion gel. (E) During the application of a large number of consecutive negative presynaptic spikes, anions accumulate near the ONW and some anions penetrate the PEO and P3HT; this movement significantly increases charge carrier density in the ONW. (F) After the presynaptic spikes, the accumulated anions gradually drift back to equilibrium positions, whereas the spontaneous release of the trapped anions in the ONW is slow, inducing long-term plasticity.

5 E. Control over spacing of ONWs Sets of parallel nanowires with spacings of 2, 2.5, 5, 7.5, 10, 15, 20, 30 and 50 m were fabricated (fig. S5). fig. S5. Control over inter-onw spacing. SEM images of an ONW pair with computer-digitalcontrolled spacing of (A) 2 m, (B) 2.5 m, (C) 5 m, (D) 7.5 m, (E) 10 m, (F) 15 m, (G) 25 m, (H) 30 m and (I) 50 m.

6 F. Control over diameter of ONWs By controlling the concentration of the solution, diameters of the ONWs can be varied (fig. S6). This technique avoids the use of high-vacuum processes and allows the use of inexpensive equipment. The e-nw printing process is a promising technique due to its low cost, high speed, and computer-controllable design on a large-scale, and due to the high flexibility and transparency of the various nanowires. fig. S6. Control over ONW diameter. SEM images of an ONW with diameter of (A) 70 nm, (B) 90 nm, (C) 120 nm, (D) 150 nm, (E) 200 nm and (F) 300 nm.

7 G. Printing of other types of semiconducting polymer nanowires E-NW printing can print ONWs of other types of polymer semiconductor materials. A SEM image of aligned N2200/PVK 7:3 organic nanowires is provided as an example (fig. S7). fig. S7. SEM image of aligned N2200/PVK (7:3) polymer semiconductor NWs.

8 H. Short-term synaptic enhancement fig. S8. Short-term synaptic enhancement. (A) Schematic of applying sequential electrical pulses with time interval tpre analogous to presynaptic spikes to an ONW ST that induces current responses through the ONW active channel. (B) Short-term synaptic enhancement = A2/A1 (A1 and A2 are labeled in Fig. 2E) vs tpre between the two consecutive pre-synaptic spikes.

9 I. Curve of forgetting The current-retention behavior for this ion accumulation mechanism could be conceived as analogous to memory retention in a human brain, in that the spontaneous decay of the EPSC is similar to that of forgetting curve, which shows how information is lost over time when no attempt is made to retain it (37). The current-retention behaviors fit very well with the most used equation for forgetting curve y = b t -m (fig. S9), where y is memory retention, t is time, and b and m are empirically-fitted parameters (38). fig. S9. Current-retention behavior fits well with the curve of forgetting. (A) EPSC triggered by one pulse (black dots) and fitted curve of forgetting (red line), (B) EPSC triggered by two pulses (black dots) and fitted curve of forgetting (red line) and (C) EPSC triggered by three pulses (black dots) and fitted curve of forgetting (red line).

10 J. Amplitude of the depression Figure 2D shows IPSC with 16.5% decrease from the resting current; this decrease appears to be a reasonable value in response to a single presynaptic spike. To more emphatically demonstrate IPSC, additional spikes were applied (fig. S10). IPSC that was triggered by the positive presynaptic spikes was reduced from the resting current by 16.5% (one pulse), 30.8% (two pulses), 43.5% (three pulses) and 74.2% (10 pulses). Post-synaptic current (A) 2.7n 1.8n 900p 16.5% 30.8% 43.5% % (10 pulses) Time (s) fig. S10. Further decreased IPSC was observed with an increased number of presynaptic spikes. IPSC triggered by one pulse, two pulses, three pulses and ten pulses were reduced by 16.5%, 30.8%, 43.5% and 74.2% from the resting current level, respectively.

11 K. Spike timing dependent plasticity Spike timing dependent plasticity (STDP), which describes the change of synaptic weight in response to the relative timing of the spikes of pre- and post-neurons, is an important synaptic adaption rule of Hebbian learning that is often summarized as "Cells that fire together, wire together". As with other types of the synaptic plasticity, it is widely believed to underlie information processing and storage in the brain. A typical asymmetric form of STDP (39) induced by temporal correlations of pre- and post-synaptic spikes (fig. S11A) was obtained (fig. S11B) (40,41). The pulse shape, amplitudes and time duration of pre- and post-synaptic spikes have been provided in fig. S11A. fig. S11. Spike timing dependent plasticity. (A) Schematic of application of presynaptic spikes and postsynaptic spikes to a synaptic joint. (B) STDP of the ONW artificial synapse as a demonstration of Hebbian learning rule, with synaptic weight change plotted as a function of the relative timing of pre and postsynaptic spikes ( tpost-pre).

12 L. Nanogap formation fig. S12. SEM images for the nanogap formation. (A) PVK nanowire. (B) Deposition of Ti/Au. (C) Nanogap formation. (D) 10 ONWs were printed across the nanogap.

13 M. Some discussion of energy efficiency The low energy consumption of the devices derives from the small feature size and core/sheath structure of the ONW (fig. S3), and the short channel length. The ONWs had a diameter ~200 nm (Fig. 4C, inset) and a core/sheath structure with a P3HT inner core in a PEO sheath (figs. S3 and S13). Their nano-size fiber-like structure permits large surface area-to-volume ratio that improves interfacial contact between the active channel and ion gel; this contact is critical for absorption and release of ions. Increasing the number of absorbed ions increases the number of charge carriers in the active channel correspondingly; this is consistent with results in previous ion gel-gated transistors in which enlarged contact area significantly increased the gating effect and mobility by increasing the number of pathways along which ions could penetrate the P3HT channel to induce electrochemical doping that significantly increases charge carrier density (19,42); this increase may allow reduction in operating voltage and in power consumption. The extremely short channel length reduces the path of energy dissipation, and significantly increases energy efficiency. The reduction in channel length reduces energy dicipation, which has been previously suggested (34) but experimentally realized in this paper. The key factors that result in this extemely low energy consumption are (1) the very large surface-to-volume ratio of the ONW and (2) the very short channel length that significantly shortens the path along which energy can dissipate. The ion gel is ionically conductive and electrically insulating (19) with leakage current magnitudes smaller than that of the semiconducting path in similar transistor geometries, so energy dissipation through the ionic gating is negligible (35). fig. S13. Configuration of a typical ion gel gated ONW ST. The ONW ST composed of ion gel, P3HT/PEO core-sheath structured ONW, substrate, and a metal probe. The ion and charge carrier distribution mimic the moment at which the negative presynaptic spike was applied by the metal probe.

14 N. Spike-rate dependent plasticity of short-channel device The change in postsynaptic current intensity as a function of spike rate is an important feature of synapses. The short-channel synaptic transistor showed this plasticity with low energy consumption (fig. S14). -8p Post-synaptic current (A) -6p -4p Time (s) fig. S14. Spike rate dependent plasticity of a short-channel ST. EPSC peaks triggered by 10 presynaptic spikes with the same spike amplitude (-1 mv) and duration (300 ms), but different spike rates: 1200 ms/spike, 1500 ms/spike and 1800 ms/spike.

15 O. Presynaptic spike-voltage-dependent EPSC 300n -1 V -2 V -10 V EPSC (A) 200n 100n Pulse width fixed at 50 ms Time (s) fig. S15. Presynaptic spike voltage dependent EPSC behavior of the ONW ST. P. Presynaptic spike-duration-dependent EPSC 150n 50ms 100ms 500 ms 100n Pulse intensity fixed at -1 V EPSC (A) 50n Time (s) fig. S16. Presynaptic spike duration dependent EPSC behavior of the ONW ST.

16 Q. Narrow pulse triggered EPSC Mammalian cells express more than a dozen different types of ion channels to encode information by generating action potentials with a wide range of shapes, frequencies and patterns. Different neurons can produce diverse types of action potentials, which also vary in different stage of firing. To show sensitivity of the device, we use an 8-ms pulse width that is in the reasonable range of dopaminergic neurons in central brain (43) to trigger EPSC and IPSC in a short-channel ONW ST (fig. S17). The physical limit of the Keithley 4200 restricts the pulse width to a minimum of 8 ms. fig. S17. Postsynaptic current in ONW ST triggered by 8-ms pulses. (A) EPSC triggered by one negative pulse. (B) IPSC triggered by one positive pulse. (C) EPSC triggered by ten negative pulses. (D) IPSC triggered by ten positive pulses.

17 R. Calculation of signal-to-noise ratio Signal-to-noise ratio (SNR) is roughly estimated using the typical EPSC signal in Fig. 2C, according to the following equation SNR = P S = r 2 S P N where, PS is the power of a signal and PN is the power of background noise, rs is the distance from signal peak to the average baseline, and 2 is the variance of the noise. The signal amplitude r (fig. S18A) and 20 points of base line (fig. S18B) for calculation of their variance were selected from Fig. 2C. The calculated SNR was , which can be converted according to the following relationship to 52.6 db. N 2 SNR (db) = 10log 10 P S P N

18 fig. S18. Data selection for SNR calculation. (A) The selection of r from Fig. 2C. (B) The selection of discrete point from the baseline for the calculation of standard deviation, where N 2 = 1 N (I N I avg ) 2.

19 S. Transistor characteristics of the device fig. S19. Transistor characteristics of the electronic devices. (A) Transfer curve of the electronic device. (B) Output curve of the electronic device. The device is composed of 5 parallel ONWs across a single channel (Channel Length = 50 m).