High-Speed Digital Design and Verification
Transcription
High-Speed Digital Design and Verification
High-Speed Digital Design and Verification Russ McHugh Senior Application Engineer, High Speed Digital and Analog Test How to characterize and debug high speed digital links on your physical prototype? What part of your design is eating up your Eye margins? Signal Integrity Challenges TP0 Tx - EQ TXn TP2 Channel TP3 Connector + Connector TXp TP1 RXp + RXn Inter-Symbol Interference (ISI) Probing Effects Impedance Mismatches and Reflections Cross Talk Loss Methods to Compensate for Measurements Calibration Mathematical Signal Processing De-embedding/Embedding Techniques 2 TP4 EQ Rx - Separating out the Transmitter Effects TX Channel RX Capturing an eye diagram The easiest way to get an overall idea of the quality of the serial signal Created relative to a clock (explicit, recovered) Eye Diagram is the superposition of all the bit transitions Clock Recovery settings greatly effect the shape of the eye 101 Sequence 011 Sequence Overlay of all combinations What represents “good enough”? The Eye Mask The eye-mask is the common industry approach to measure the eye opening Failures usually occur at mask corners • But what is cause of failure? Violating USB FS 12Mb/s Eye Diagram Good Displayport Eye Diagram Types of Jitter Jitter Bounded UnBounded Deterministic (DJ) Correlated with Data (DDJ) Uncorrelated with Data Duty Cycle Distortion Inter Symbol Interference (DCD) (ISI) Tr, Tf D BW Limits Xtalk Reflections Non Linear CR Tx Threshold Random (RJ) Aperiodic Periodic (ABUJ) (PJ) Events Clocks Uncorrelated with Data Skewed Gaussian Thermal Shot 1/f Burst Measuring Jitter: Bit Error Ratio (BER) Testing The only way to directly measure Total Jitter is with a Bit Error Ratio (BER) test. Sample at various points along unit interval, directly measure BER at each point. Plot “bathtub” curve. 0.5 BER 10-3 Gaussian Tails 10-6 10-9 Eye Opening at BER=10-12 10-12 W 0 0.5TB TB /UI Can be slow depending on the data rate TJ(BER) = UI – W Time Interval Error (a) Clock Reference time (b) Source Waveform voltage measurement threshold 0 time time error (c) Time Interval Error (TIE) 0 Measures total jitter of the acquisition. The more transitions you measure, the greater TIE will become. time Waveform transitions deviate from expected transition time Generate Time Interval Error (TIE) by measuring transitions versus reference clock Dual Dirac Model – Total Jitter BER n 1x10-8 5.73 1x10-10 6.47 1x10-12 7.13 1x10-14 7.74 L [ ( x L ) Total Jitter n= R + ( x R )] TJ(BER) = RJpp / 2 DJ JDJ PP ( = RJrms) x2 exp 2 2 UI – W 7 L R (x L )2 (x R )2 exp expdirectly 2 TJ can be measured 2 2 2 using a BER test OR TJ(BER) = 2n*RJrms + DJ TJ can be calculated using RJ and DJ in the Dual Dirac model Tail Fit Problem Total Jitter Histogram RJ/PJ/ABUJ Histogram 1 The spectral method lumps ABUJ in with RJ and greatly overestimates TJ. 0.9 0.8 0.7 Solution 0.6 0.5 0.4 1. Perform Gaussian Tail fit on RJ/PJ/ABUJ Histogram to get RJrms directly RJrms 0.3 0.2 2. Spectral fit gives (RJ+ABUJ)rms & PJrms 0.1 0 -30 -20 -10 0 10 20 30 3. Subtract RJ rms to get ABUJrms EZJIT Plus – Advanced Jitter Decomposition Easy to use wizard guides you through jitter measurement setup Fully compatible with Infiniium Software such as Equalization, PrecisionProbe, and InfiniiSim Customizable jitter views Jitter Measurement Demonstration Page 12 Why is vertical noise floor important ? Let’s consider theoretical signals with Zero jitter and fixed voltage noise with three different edge speeds and crossing a Threshold at 50% 1)Voltage noise translates directly to Timing uncertainty (Jitter) 2)Higher Vertical Noise Floor translates to higher Timing uncertainty 3)At constant amplitude noise floor, slower edge speed translates to higher timing uncertainty EZJIT Complete – Vertical Noise Decomposition Vertical slice of real time eye allows for noise decomposition using the same concepts and algorithms as timing jitter Analyze One level, Zero level, or both Vertical Noise Measurement Demonstration Page 15 Separating out the Channel Effects TX Channel RX Analyzing a serial Link TX Clean Source Signal Channel Channel Frequency Response We are going to analyze a 12Gb/s Link Channel will be 9 Inch FR4 PCB Page 17 RX Closed Eye Received Signal Eye & Jitter Break Down Analysis on TX output Transmitter 12Gb/s Intrinsic Jitter Analysis 33GHz 80GSa/s Scope RJ: 500fs (RMS) PJ: 740fs DCD: 660fs ISI: 10.52ps AGILENT SI Seminar 2012 by Pascal GRISON Page 18 Eye Diagram on TX output and Channel Output Depending on Link Target Data rate & Transmission Channel Losses AGILENT SI Seminar 2012 by Pascal GRISON AGILENT SI Seminar 2012 by Pascal GRISON Even with Perfect TX Eye Opening… You may end up with a completely closed at Receiver Side Why is the RX Eye Closed? ISI Jitter! Does that mean that this link will never Work? Well it Depends…. Page 19 What is Inter-Symbol Interference? AGILENT SI Seminar 2012 by Pascal GRISON ISI Jitter is coming from Signal Distortions in Transmission Channel Page 20 Scope can Emulate Receiver EQUALIZATION Modern SerDes are incorporating Reciever Equalization Using Oscilloscope Equalization we can emulate most DUT RX EQ configurations: AGILENT SI Seminar 2012 by Pascal GRISON • Feed Forward Equalizer Continuous Time Linear Equalizer • Decision Feedback Equalizer Page 21 Emulate Receiver EQUALIZATION on Oscilloscope From almost Zero RX Eye Opening with no TX DeEmphasis and No RX EQ AGILENT SI Seminar 2012 by Pascal GRISON RX Eye Opening of 132mV X 65ps Was achieved with EQUALIZATION You MUST Emulate your RX Equalization in Oscilloscope to Analyze True RX Eye Diagram Page 22 Commonly Encountered Insertion Losses Typical Insertion Loss Through FR4 100MHz …….. 2 dB/m (20%/m) 500MHz …….. 4 dB/m (37%/m) 1GHz ………… 7 dB/m (55%/m) 5GHz ………… 25 dB/m (94%/m) 10GHz ……….. 45 dB/m or 1.14 dB/in (99%/m) 100MHz …….. 1GHz ………… 5GHz ………... 10GHz ………. 20GHz ………. Typical Insertion Loss Through SMA 0.25 dB/m (2.8%/m) 0.8 dB/m (8.8%/m) 2 dB/m (20%/m) 3 dB/m (29%/m) 5 dB/m (44%/m) Loss increases exponentially with frequency. Non-idealities you could ignore at <1GHz now severely erode design margins Same Measurement Setup, VERY Different Results Corrected Uncorrected • PCI Express 2, SATA, and Custom • Compliance Requirement for Gen 2 Tx PHY 25 Channel S4P Connector • Compensate for Probing and Fixture Loss – Add Margin to Transmitter Characterization Connector De-embedding – Loss Compensation or Gain Function (De-convolve) Rx PHY De-embedding the DDR2 BGA Probe 26 Probe at VIA De-embed Probe Probe at BGA RT BGA = 390 ps RT VIA = 183 ps RT De-embed = 175 ps 27 Embedding – Loss Function (Convolve) Pin Rx TP1 TP2 TP3 Virtual Probe and Deemphasis/Equalization • Simulate Equalization/De-emphasis at Rx Connector • Simulate Channel Loss on Signal Measured at Tx Tx PHY Tx Signal Virtual Probe Rx Equalization TP1 TP3 TP2 28 Channel Connector Connector Channel.s4p+ conn.s4p+package.s4p Rx PHY Separating out the Receiver Effects TX Channel RX How do we ensure ‘Error Free’ operation? How can we specify to the transmitter plus channel design the quality of the signal that the receiver needs? Answer - We need to measure the Receiver’s tolerance to jitter. 30 Characteristic eye closure by sinusoidal Jitter Eye Diagram BER Scan 31 Receiver Jitter Tolerance with Loopback JBERT up to 28Gb/s PRBS Generation with Calibrated Jitter insertion and integrated adjustable ISI channel DUT SerDes in Loopback Mode RX Data Receiver Rx latch AGILENT SI Seminar 2012 by Pascal GRISON ISI Channel DLL Rx PLL Transmitter Tx latch JBERT Real time Error Detector allow thorough BER Analysis Tx DLL TX Data Customers are using J-BERT to characterize SERDES receiver susceptibility to ISI, Random Jitter, Periodic Jitter, and BUJ 32 RX Testing with asycronous devices (SER / FER Support) Error detector supports 8B/10B encoded data, and can deal with re-timed data and filler symbols Data-out with fBERT Re-timed data with fProduct D+ D- D+ F+ F+ D+ D- D+ D+ D- D+ F+ D+ D- D+ EQ Clock FF CDR Elastic buffer Receiver CDR Loopback Transmitter Product under test Expected data with fproduct D D D D Device function D D Loopback data with fproduct and new disparity D- D+ D- F- D- D+ D- Clock Rx Compliance and Jitter Tolerance Testing 34 Summary • TX characterization is frequently captured with eye diagrams. • The Dual Dirac model allows us to extrapolate jitter measurements to low BER rates. • Amplitude noise can contribute to random jitter. • ISI is a dominant source of eye closure at high data rates. • De-embedding and virtual probing can correct for fixture and provide measurements at inaccessible locations. • Equalizers are a key feature of receivers. • Receiver jitter tolerance testing is becoming a requirement. • Asynchronous systems require the BER error detector to handle filler symbols. Page 36 Questions Page 37