Browsing by Author "Kuo, J. L."

Now showing 1 - 2 of 2

An Energy Efficient FinFET-based Field Programmable Synapse Array (FPSA) Feasible for One-shot Learning on EDGE AI
(2018-01-01)
Kuo, J. L.
;
Chen, H. W.
;
Hsieh, E. R.
;
Chung, Steve S.
;
Chen, T. P.
A pure logic 14nm FinFET with capabilities of linearly tunable V-th and excellent retention has been implemented as synapses in neuromorphic system., a Field Programmable Synapse Array (FPSA) has been adopted to replace conventional R-based memory Synapse Array (RSA). Thanks to the wide range of V-t-tuning ability, 200X on/off ratio, and the ultra-small variability, 12%, results showed that the training power and SN ratio of FPSA are 10 times and 50 times smaller than those of the RSA, respectively. Two applications were demonstrated on FPSA array for one-shot learning applications. First, FPSA is used to detect handwritten digits of MNIST dataset. "Learned it by once" can be achieved in this task. Furthermore, FPSA has been applied to recognize goldfish in Cifar 100 dataset after learned the other 4 fish species. With the assistance from one-shot learning, results show the machine learned it faster and better on EDGE. This demonstrates the feasibility of FPSA for low-power and cost-effective synapse-based one-shot learning applications in the AIoT era.
Quantum-mechanical approach to predissociation of water dimers in the vibrational adiabatic representation: Importance of channel interactions
(2015-08-28)
Mineo, H.
;
Niu, Y. L.
;
Kuo, J. L.
;
Lin, S. H.
;
Fujimura, Y.
The results of application of the quantum-mechanical adiabatic theory to vibrational predissociation (VPD) of water dimers, (H2O)(2) and (D2O)(2), are presented. We consider the VPD processes including the totally symmetric OH mode of the dimer and the bending mode of the fragment. The VPD in the adiabatic representation is induced by breakdown of the vibrational adiabatic approximation, and two types of nonadiabatic coupling matrix elements are involved: one provides the VPD induced by the low-frequency dissociation mode and the other provides the VPD through channel interactions induced by the low-frequency modes. The VPD rate constants were calculated using the Fermi golden rule expression. A closed form for the nonadiabatic transition matrix element between the discrete and continuum states was derived in the Morse potential model. All of the parameters used were obtained from the potential surfaces of the water dimers, which were calculated by the density functional theory procedures. The VPD rate constants for the two processes were calculated in the non-Condon scheme beyond the so-called Condon approximation. The channel interactions in and between the initial and final states were taken into account, and those are found to increase the VPD rates by 3(1) orders of magnitude for the VPD processes in (H2O)(2) ((D2O)(2)). The fraction of the bending-excited donor fragments is larger than that of the bending-excited acceptor fragments. The results obtained by quantum-mechanical approach are compared with both experimental and quasi-classical trajectory calculation results. (C) 2015 AIP Publishing LLC.