This document describes the modeling methodology used to assess the performance of these probes in terms of allowed digital bandwidth of signals chosen for temporary fault insertion trials. This methodology is based on time-domain characterization of Scattering parameters (TDR/TDT) and subsequent extraction of a Behavioral Time-domain Model (BTM) [13] of the probe itself. This technique called PWLFIT (Piece-Wise Linear FITting) [14] [15]is supported by the Digital Wave Simulator DWS [16] [17] and its companion tool DWV [18] developed starting in the early '90s for very fast modeling and simulation of high-speed circuits and systems.
Driving Behavioral Change for Information Management through Data-Driven Gree...
Multigigabit modeling of hi safe+ flying probe fp011
1. 1 Piero Belforte July 17 2018
MULTIGIGABIT MODELING OF A HiSAFE+ DIFFERENTIAL FLYING PROBE
Introduction
Telecom equipments are characterized by stringent fault tolerance constraints to comply to
service availability targets . This has been true from the very beginning of the digital era of Public
Switched Telecom Networks (PSTNs) in the years 70's [1] [2] [3] [4] and then of the Internet era
started in the years 90's. Network nodes require redundant structures to limit the effects of
faults on the quality of service and suitable testing procedures are required to test the diagnostics
and spare units reconfiguration. Among the available methodologies, hardware Fault Insertion
Testing (FIT) and FMEA (Fault Modes and Effects Analysis) are among the most relevant ones.
In the years '90 at CSELT Labs [5] of Turin, Italy, a specific project called THRIS [6] [7] [8] [9] was
started to develop new reliable FIT methods to be applied to telecom devices. The developed
THRIS instrument was then extensively used to improve the reliability of the new digital switches
to be deployed in the network. At the beginning of years 2000 Cisco Systems required a more
performing FIT equipment able to inject controlled hardware faults inside high-end IP routers.
HiSAFE [10] , the new high-speed FI tester was specifically designed to fulfill these requirements
by means of new fault insertion methodology. In 2013 a new higher speed version, HiSAFE+ [11],
was developed by using new larger bandwidth probes [12]. More recently (2018) new flying
probes have been designed to fulfill the requirement of FMEA testing of new generation IP
routers.
This document describes the modeling methodology used to assess the performance of these
probes in terms of allowed digital bandwidth of signals chosen for temporary fault insertion trials.
This methodology is based on time-domain characterization of Scattering parameters (TDR/TDT)
and subsequent extraction of a Behavioral Time-domain Model (BTM) [13] of the probe itself. This
technique called PWLFIT (Piece-Wise Linear FITting) [14] [15]is supported by the Digital Wave
Simulator DWS [16] [17] and its companion tool DWV [18] developed starting in the early '90s for
very fast modeling and simulation of high-speed circuits and systems.
2. 2 Piero Belforte July 17 2018
PWLFIT model of a HiSAFE+ FP011 differential flying probe
A behavioral model (BTM) of a prototype of the differential probe FP011 is built up by extraction
of of pwl breakpoints using the MCS utility included in DWV. The source waveform is the
unbalanced (Ch1) response of the probe connected to a CSA803C TDR in a differential
configuration by means of a suitable fixture including the semi-rigid launch cables (Fig. 1,2). The
Ch1 waveform of the fixture without the FP011 probe is also acquired to be used as input edge in
the de-embedding setup (Fig. 3,4).
Fig. 1: CSA803C differential setup including the fixture for the FP011 probe
Fig.2: Measured reflection (S11) related to the configuration shown in Fig.1 (channel 1) and its
extracted pwl model.
3. 3 Piero Belforte July 17 2018
Fig. 3: CSA803C differential setup to acquire the edge at the fixture without the probe
Fig.4: Measured edge related to the configuration shown in Fig.3 (no probe,channel 1) and its
extracted pwl model.
A specific DWS simulation setup is built up in order to accomplish the step-by-step optimization
procedure (Fig.5).This configuration is used to match the reflected wave of the probe model at
port 20, B(FP011P,20), with the target waveform V(40) by changing the breakpoints of the pwl
model with respect the initial guess of the netlist shown in Fig.6. This is done by manual editing
the breakpoints values of the element FP011P in and observing the result of the new simulation
run. This process is repeated until a satisfactory match between the behavioral probe model and
the target is obtained. The result of the initial step is shown in Fig.7 where the differences
between the model response and the target are outlined in yellow. The damping of the first
4. 4 Piero Belforte July 17 2018
oscillation is due to the bandwidth limitation of the incident edge. For this reason the first
breakpoints are modified in order to enhance the amplitude of the reflection peak in the first 30ps
of the response. This reflection peak is mainly due to the stub effect of the 3mm-long probe tips.
Only a few steps are required to get a satisfactory result. The modified netlist at the final step is
shown in Fig.8. Only 3 breakpoints are needed to describe the reflection peak due to the inductive
stub effect of the probe tips (Fig. 9,10). The final match to the target is shown in Fig. 11 where the
residual error is outlined in yellow.
Fig.5: Schematic of the configuration used to match the reflected wave of the probe model
B(FP011P,20) with the target waveform V(40) by changing the breakpoints of the pwl model.
5. 5 Piero Belforte July 17 2018
Fig.6: DWS netlist related to the schematic of the configuration of Fig. 5. Here the initial step is
shown where the breakpoints of the actual response are used as initial guess.
6. 6 Piero Belforte July 17 2018
Fig.7: Simulated waveforms related to the initial DWS netlist of Fig. 5. Here the initial step is
shown where the breakpoints of the actual response are used as initial guess. The error between
the reflected wave and the target is outlined in yellow.
7. 7 Piero Belforte July 17 2018
Fig.8: DWS netlist related to the schematic of the configuration of Fig. 5. Here the final step is
shown where the first breakpoints of the actual response are modified in order to match the
actual response.
8. 8 Piero Belforte July 17 2018
Fig.9: Simulated waveforms related to the DWS netlist of Fig. 8. Here the final optimization step is
shown where the initial breakpoints of the actual response have been modified to match the
model with the actual response. The error between the reflected wave and the target is outlined
in yellow.
Fig.10: Simulated waveforms related to the DWS netlist of Fig. 8. Here the result of the final
optimization step is shown. The difference between the de-embedded pwl model and the initial
guess is outlined in yellow.
9. 9 Piero Belforte July 17 2018
Fig.11: Simulated waveforms related to the DWS netlist of Fig. 8. Here the result of the final
optimization step is shown in terms of final difference between the target and de-embedded
model response. This residual error is shown in yellow.
Digital bandwidth assessment of the probe
The most effective way to assess the bandwidth performance of a digital device or system is to
evaluate its worst-case eye-diagram (WCED) contour at various bit-rates. This task can be easily
accomplished by using the specific limit-eye utility included in DWV. The result is shown in Fig.12
where the 2,5,10Gbs WCEDs related to the de-embedded response of FP011 probe (Fig. 10) are
shown superimposed using different colors. It can be easily noticed that a 50ps long closure
affects the contour at all bit-rates considered, while for the rest of the bit-time the contours are
well open. As previously pointed up this is mainly due to the inductive stub effect of the probe
tips and can be reduced only by adopting shorter tips. At 2 Gbps this effect is negligible because
affects only a 10% of the bit time. At 5 Gbps it affects a 25% of the bit-time, while at 10Gbps only a
50% of the contour is still open. In practice this means that probe FP011 can be used up to 5Gbps
to inject faults in high-speed differential nets (signals) of the system under test.
10. 10 Piero Belforte July 17 2018
Fig.12: Worst-case inner eye contours related to the de-embedded FP011 probe model at 2,5 and
10 Gbps.
Concluding remarks
A very fast model extraction methodology (PWLFIT) has been successfully applied to the new
differential flying probe FP011 (Fig. 13) of the HiSAFE+ Fault Insertion Tester. The assessed digital
bandwidth of the probe is about 5Gbps, covering the needs of reliable FMEA tests on new
generation IP routers.
A similar method has been applied to model single-ended version of flying probes (FP001+ and
FP002+) capable of inject fault on nets (signals) up to 2 Gbps [19] [20].
The extracted BTM models can be used for fast and reliable DWS simulations of the net under test
in all operating situations (probe connected in off or on conditions ) to check for both signal
integrity (off) and stuck-at reliable fault injection (on).
Larger bandwidth (up to 20 Gbps) can be achieved by using differential solderable probes like
model 010 and model 012 provided within the HiSAFE+ tester.
PWLFIT method can become fully automatic by future developments of suitable pwl fitting
algorithms.
11. 11 Piero Belforte July 17 2018
Fig.13: Detail of flying probe FP011 prototype connected to a SERDES net vias.