We demonstrated a proton-based 3-terminal synapse device which shows symmetric conductance change characteristics. Using the optimized device, we successfully confirmed the improved classification accuracy of neural networks for on-chip training.

A second electron cyclotron emission imaging (ECEI) system has been installed on the KSTAR tokamak, toroidally separated by 1/16th of the torus from the first ECEI system. For the first time, the dynamical evolutions of MHD instabilities from the plasma core to the edge have been visualized in quasi-3D for a wide range of the KSTAR operation (B0 = 1.7∼3.5 T). This flexible diagnostic capability has been realized by substantial improvements in large-aperture quasi-optical microwave components including the development of broad-band polarization rotators for imaging of the fundamental ordinary ECE as well as the usual 2nd harmonic extraordinary ECE.

Since the first H-mode discharges in 2010, the duration of the H-mode state has been extended and a significantly wider operational window of plasma parameters has been attained. Using a second neutral beam (NB) source and improved tuning of equilibrium configuration with real-time plasma control, a stored energy of W tot ∼ 450 kJ has been achieved with a corresponding energy confinement time of τ E ∼ 163 ms. Recent discharges, produced in the fall of 2012, have reached plasma β N up to 2.9 and surpassed the n = 1 ideal no-wall stability limit computed for H-mode pressure profiles, which is one of the key threshold parameters defining advanced tokamak operation. Typical H-mode discharges were operated with a plasma current of 600 kA at a toroidal magnetic field B T = 2 T. L-H transitions were obtained with 0.8-3.0 MW of NB injection power in both single-and double-null configurations, with H-mode durations up to ∼15 s at 600 kA of plasma current. The measured power threshold as a function of lineaveraged density showed a roll-over with a minimum value of ∼0.8 MW at ne ∼ 2×10 19 m −3 . Several edge-localized mode (ELM) control techniques during H-mode were examined with successful results including resonant magnetic perturbation, supersonic molecular beam injection (SMBI), vertical jogging and electron cyclotron current drive injection into the pedestal region. We observed various ELM responses, i.e. suppression or mitigation, depending on the relative phase of in-vessel control coil currents. In particular, with the 90 • phase of the n = 1 RMP as the most resonant configuration, a complete suppression of type-I ELMs was demonstrated. In addition, fast vertical jogging of the plasma column was also observed to be effective in ELM pace-making. SMBI-mitigated ELMs, a state of mitigated ELMs, were sustained for a few tens of ELM periods. A simple cellular automata ('sand-pile') model predicted that shallow deposition near the pedestal foot induced small-sized high-frequency ELMs, leading to the mitigation of large ELMs. In addition to the ELM control experiments, various physics topics were explored focusing on ITER-relevant physics issues such as the alteration of toroidal rotation caused by both electron cyclotron resonance heating (ECRH) and externally applied 3D fields, and the observed rotation drop by ECRH in NB-heated plasmas was investigated in terms of either a reversal of the turbulence-driven residual stress due to the transition of ion temperature gradient to trapped electron mode turbulence or neoclassical toroidal viscosity (NTV) torque by the internal kink mode. The suppression of runaway electrons using massive gas injection of deuterium showed that runaway electrons were avoided only below 3 T in KSTAR. Operation in 2013 is expected to routinely exceed the n = 1 ideal MHD no-wall stability boundary in the long-pulse H-mode ( 10 s) by applying real-time shaping control, enabling n = 1 resistive wall mode active control studies. In addition, intensive works for ELM mitigation, ELM dynamics, toroidal ro...

A new and more accurate technique is presented for determining the toroidal mode number n of edge-localized modes (ELMs) using two independent electron cyclotron emission imaging (ECEI) systems in the Korea Superconducting Tokamak Advanced Research (KSTAR) device. The technique involves the measurement of the poloidal spacing between adjacent ELM filaments, and of the pitch angle α* of filaments at the plasma outboard midplane. Equilibrium reconstruction verifies that α* is nearly constant and thus well-defined at the midplane edge. Estimates of n obtained using two ECEI systems agree well with n measured by the conventional technique employing an array of Mirnov coils.

Temporal and spatial modulation of the edge localized mode (ELM) structure has been observed during the inter-ELM-crash period by toroidally-separated two electron cyclotron emission imaging systems. The observed modulation is interpreted as a beat wave of two modes with adjacent toroidal mode number. An additional assumption is that each mode has to have a different poloidal rotation speed. In nonlinear simulation, the low-n mode can be driven by locking between the dominant modes. The modulation is reconstructed using beat waves not the locking of modes.