Somha, Worawit

Loading...
Thumbnail Image
Preferred name
Somha, Worawit
Main Affiliation
Email
worawit.so@kmitl.ac.th

Settings

Now showing 1 - 4 of 4
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    A noise suppressing filter design for reducing deconvolution error of both-directions downward sloped asymmeric RTN long-tail distributions
    (2016-04-19)
    Yamauchi, Hiroyuki
    ;
    Somha, Worawit
    A noise suppressing filter design technique to reduce deconvolution error of both-directions downward sloped asymmetrical long-tail distribution of the Random Telegraph Noise (RTN) is proposed. The filter is used in Lucy-Richardson-deconvolution (LRDec) iteration process. The deconvolution is required for inversely analyzing RTN long tail distribution effects on VLSI time-dependent operating margin. The proposed noise suppressing filters avoid unwanted phase misalignment between the distribution curves of feedback gain and deconvolution target for right and left tails. This results in reduction of its relative deconvolution errors by about 12-fold compared with the conventional LRDec. The accuracy of the fail-bit-count (FBC) prediction is increased by about 100-fold.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    A filter design to increase accuracy of Lucy-Richardson deconvolution for analyzing RTN mixtures effects on VLSI reliability margin
    (2016-02-12)
    Yamauchi, Hiroyuki
    ;
    Somha, Worawit
    ;
    Song, Yuan Qiang
    A filter design to improve convergence characteristics in the Lucy-Richardson-deconvolution (LRDec) iterations is proposed, which is required for inversely analyzing log-mixtures 7-segmented Random Telegraph Noise (RTN) distribution effects on VLSI reliability margin. The proposed filter alleviates unwanted phase misalignment between the two distribution curves of feedback gain and deconvoluted RTN. This contributes to reduce its relative deconvolution errors by 1.5-orders of magnitude compared with the conventional LRDec. The accuracy of the fail-bit-count (FBC) prediction is increased by 10-folds while accelerating its convergence speed by 7 times of the conventional one. This contributes not to give up on a benefit of smaller iteration cycles from LRDec.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Feedback gain phase alignment effects on convergence characteristics in Lucy-Richardson deconvolution for inversely predicting complex-shaped RTN distributions
    (2015-09-28)
    Yamauchi, Hiroyuki
    ;
    Somha, Worawit
    A technique to prevent deconvolution object from ringing errors in Lucy-Richardson-deconvolution (LRDec) iteration cycles is proposed, which is required for inversely analyzing complex-shaped Random Telegraph Noise (RTN) effects on SRAM margin variations. The proposed feedback gain phase alignment successfully circumvents the unwanted errors and makes it possible to avoid the need to give up on a benefit (smaller iteration cycles) from LRDec. The ringing elimination in a real LRDec analysis for the complex-shaped RTN distribution comprising three different sloped segments has been demonstrated for the first time, while exploiting a quicker convergence benefit of LRDec algorithm. The proposed technique reduces its relative errors of the complex-shaped RTN deconvolution by about 10 times, compared with the conventional LRDec.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Ringing error prevention techniques in Lucy-Richardson deconvolution process for SRAM space-time margin variation effect screening designs
    (2015-05-05)
    Yamauchi, Hiroyuki
    ;
    Somha, Worawit
    This paper proposes a ringing error avoidance technique in Lucy-Richardson-deconvolution (L-R-Dcnv) process, which is used for inversely analyzing the Random Telegraph Noise (RTN) effects on overall SRAM margin variations. The proposed ringing prevention technique successfully circumvents the ringing error by reducing the phase difference between the feedback-gain and deconvolution target distributions in L-R-Dcnv iteration cycles. This avoids any unwanted positive feedbacks, resulting in no error amplification. This effectiveness has been demonstrated with applying it to a real L-R-Dcnv analysis for the effects of the RTN on the overall SRAM margin variations, while exploiting a quicker convergence benefit of L-R-Dcnv algorithm. It has been shown that the proposed technique reduces its relative errors of the RTN deconvolution by 10<sup>2</sup>~10<sup>3</sup> times compared with the conventional L-R-Dcnv. This enables to increase an accuracy of the fail-bit-count prediction based on the cumulative density function (cdf) of the convolution of the RTN with the Random Dopant Fluctuation (RDF) by over 2-orders of magnitude while accelerating its convergence speed by 7~30 times of the conventional one.