Bit Error Rate Test of DHPT 1.0 Gigabit Serial Link Leonard Germic , Carlos Marinas, Hans Krüger and Norbert Wermes germic@physik.uni-bonn.de University of Bonn today, 2014
Content Gigabit Serial Link 1 Test Setup 2 Results 3 germic@physik.uni-bonn.de 2 / 12
Gigabit Serial Link of DHPT 1.0 Facts: 1 . 6 GHz differential output programmable preemphasis off-chip 100 Ω matching resistors 1 . 2 V domain (VDD_CML, core voltage) germic@physik.uni-bonn.de 3 / 12
Why preemphasis Compensation of the attenuation of high frequency regime in the signals spectrum due to long distance transmission Increasing bandwidth Preventing single bit error Figure: Signal spectrum with preephasis. germic@physik.uni-bonn.de 4 / 12
Signal generation Preemphasis realization: Subtraction of signal S 1 with delayed signal S 2 . Two 8 bit DACs IDAC_CML_TX_BIAS and IDAC_CML_TX_BIASD One 2 bit DAC pll_cml_dly_sel S 1 ( t ) S 2 ( t ) S 2 ( t + dt ) S 1 − S 2 ( t + dt ) germic@physik.uni-bonn.de 5 / 12
Implementation Current Mode Logic Driver with programmable preemphasis Three parameters: IDAC_CML_TX_BIAS, IDAC_CML_TX_BIASD, pll_cml_dly_sel Same values of IDAC_CML_TX_BIAS and IDAC_CML_TX_BIASD do not correspond to same currents/differential voltage swing (differently sized current mirrors) germic@physik.uni-bonn.de 6 / 12
BERT Test setup 1 x Evaluation board XUPV5 − LX110t as DHH emulator 1 x Evaluation board XUPV5 − LX110t as Bit Error Rate Tester (IBERT, Xilinx) 2 x 15 m Infini band cables ... and a crazy ’handmade’ script germic@physik.uni-bonn.de 7 / 12
Figure: BERT Test system. Top board = DHH emulator, Buttom board = IBERT (Xilinx). germic@physik.uni-bonn.de 8 / 12
BERT Results - Good A and bad B case A B Figure: Red line indicates the BER thr of 10 − 6 germic@physik.uni-bonn.de 9 / 12
Optimal Working Point A B Figure: Sweep results at a threshold of BER thr = 10 − 6 . Higher is better - wider opening at BER thr germic@physik.uni-bonn.de 10 / 12
Optimal Working Point BEST Optimal working point BIAS = 15 , BIASD = 150 , dly = 0. germic@physik.uni-bonn.de 11 / 12
Thank you for your attention! germic@physik.uni-bonn.de 12 / 12
Sweep test These are NOT absolute BER values. NMOS of current mirrors saturate above DAC values ∼ 150 (Design bug) Figure: Plateau due to saturation of the current mirrors. germic@physik.uni-bonn.de 12 / 12
Recommend
More recommend