An Experimental Study of Metastability-Induced Glitching Behavior ¤ Thomas Polzer Department Electronic Engineering UAS Technikum Wien, H€ ochst € adtplatz 6, A-1200, Vienna, Austria [email protected]Florian Huemer † and Andreas Steininger ‡ Institute of Computer Engineering, TU Wien, Karlsplatz 13, A-1040 Vienna, Austria † [email protected]‡ [email protected]Received 15 October 2018 Accepted 27 March 2019 Published 9 May 2019 The increasing number of clock domain crossings in modern systems-on-chip makes the careful consideration of metastability paramount. However, the manifestation of metastability at a °ip- °op output is often unduly reduced to late transitions only, while glitches are hardly ever accounted for. In this paper we study the occurrence of glitches resulting from metastability in detail. To this end we propose a measurement circuit whose principle substantially di®ers from the conventional approach, and by that allows to reliably detect glitches. By means of exper- imental measurements on an FPGA target we can clearly identify late transitions, single glit- ches and double glitches as possible manifestations of metastability. Some of these behaviors are unexpected as they do not follow from the traditional modeling theory. We also study the dependence of metastable behavior on supply voltage. Beyond con¯rming that, as reported in previous literature, the metastable decay constant ( is voltage-dependent, we also produce strong evidence that the relative occurrence of glitches is not voltage-dependent. Keywords: Metastability; time-to-digital converter; TDC; late transition detection; carry chain. *This paper was recommended by Regional Editor Zoran Stamenkovic. ‡ Corresponding author. This is an Open Access article published by World Scienti¯c Publishing Company. It is distributed under the terms of the Creative Commons Attribution 4.0 (CC-BY) License. Further distribution of this work is permitted, provided the original work is properly cited. Journal of Circuits, Systems, and Computers Vol. 28, Suppl. 1 (2019) 1940006 (21 pages) # . c The Author(s) DOI: 10.1142/S0218126619400061 1940006-1 J CIRCUIT SYST COMP Downloaded from www.worldscientific.com by VIENNA UNIVERSITY OF TECHNOLOGY on 07/02/19. Re-use and distribution is strictly not permitted, except for Open Access articles.
21
Embed
An Experimental Study of Metastability-Induced Glitching ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
The increasing number of clock domain crossings in modern systems-on-chip makes the carefulconsideration of metastability paramount. However, the manifestation of metastability at a °ip-
°op output is often unduly reduced to late transitions only, while glitches are hardly ever
accounted for. In this paper we study the occurrence of glitches resulting from metastability in
detail. To this end we propose a measurement circuit whose principle substantially di®ers fromthe conventional approach, and by that allows to reliably detect glitches. By means of exper-
imental measurements on an FPGA target we can clearly identify late transitions, single glit-
ches and double glitches as possible manifestations of metastability. Some of these behaviors are
unexpected as they do not follow from the traditional modeling theory. We also study thedependence of metastable behavior on supply voltage. Beyond con¯rming that, as reported in
previous literature, the metastable decay constant � is voltage-dependent, we also produce
strong evidence that the relative occurrence of glitches is not voltage-dependent.
Keywords: Metastability; time-to-digital converter; TDC; late transition detection; carry chain.
*This paper was recommended by Regional Editor Zoran Stamenkovic.‡Corresponding author.
This is an Open Access article published by World Scienti¯c Publishing Company. It is distributed under
the terms of the Creative Commons Attribution 4.0 (CC-BY) License. Further distribution of this work ispermitted, provided the original work is properly cited.
Journal of Circuits, Systems, and ComputersVol. 28, Suppl. 1 (2019) 1940006 (21 pages)
the identi¯cation of double pulses ¯nally becomes possible only with our novel
measurement approach.
As a further measure to enhance the analysis of pulse behavior, we have chosen a
very asymmetric duty cycle for the clock (8.3%). As known from the literature,3 this
moves the handover from master to slave towards an earlier point in time, leaving the
master with less resolution time and thus pronouncing the metastable behavior of the
slave. By this it was possible to obtain a reasonable number of pulses within our
measurement duration of 110 h.
To motivate the speci¯c focus we give to the slave latch behavior, we want to
stress here that solely using the master's resolution constant � (which is much easier
to obtain by measurement) for the MTBU calculation can, lead to an optimistic view
of the °ip-°op's resolution capability, as this ignores the fact that during the low
phase of the clock the slave's � becomes relevant which is much worse. As already
mentioned in Sec. 1, in addition to this optimistic � selection, the occurrence of pulses
may introduce adverse e®ects into the circuit that are overlooked when considering
late transitions only.
4.3. Voltage dependence of the delay lines
To gather further insight into the pulse generation behavior of the °ip-°op, we
studied the supply voltage (Vdd) dependence. To this end, we varied Vdd between
0.98V and 1.02V in 10mV steps. The measurement duration was kept constant at
110 h. During the whole experiment the actual operating voltage at the FPGA's
supply pin was monitored with an external multimeter. The box plots in Fig. 8
visualize the distribution of deviation from the respective supply voltage set point
during the measurement. The plot whiskers extend to the most extreme data point
Fig. 8. Distribution of Vdd variation.
An Experimental Study of Metastability-Induced Glitching Behavior
1940006-15
J C
IRC
UIT
SY
ST C
OM
P D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by V
IEN
NA
UN
IVE
RSI
TY
OF
TE
CH
NO
LO
GY
on
07/0
2/19
. Re-
use
and
dist
ribu
tion
is s
tric
tly n
ot p
erm
itted
, exc
ept f
or O
pen
Acc
ess
artic
les.
that still lies in the range of three times the interquartile range. The voltage °uc-
tuations thus observed remained around 100�V, which is signi¯cantly smaller than
the 10mV stepping between measurement runs.
As supply voltage changes are known to cause changes in the delay of logic
elements, the delay line had to be calibrated for each supply voltage value separately.
Figure 10 illustrates how the observed length of the whole delay line (for both, delay
lines A and B, as well as for rising and falling transitions) depends on the supply
voltage. These results con¯rm the expected speedup of the delay line for rising Vdd.
Interestingly, the delay di®erences between rising and falling edges of the same delay
line are much more signi¯cant than those between the delay lines A and B for
matching transitions.
4.4. Supply voltage dependence of ¿
The impact of the supply voltage changes on the metastability behavior is shown in
Fig. 6. When using the relevant slopes for determining the metastability resolution
constants for master and slave latch, we arrive at Fig. 9. We observe that � decreases
signi¯cantly with increasing supply voltage. This is not unexpected, as higher voltage
(a) Master. (b) Slave.
Fig. 9. Metastability resolution constant � .
Fig. 10. Delayline length.
T. Polzer, F. Huemer & A. Steininger
1940006-16
J C
IRC
UIT
SY
ST C
OM
P D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by V
IEN
NA
UN
IVE
RSI
TY
OF
TE
CH
NO
LO
GY
on
07/0
2/19
. Re-
use
and
dist
ribu
tion
is s
tric
tly n
ot p
erm
itted
, exc
ept f
or O
pen
Acc
ess
artic
les.
(a) 01. (b) 10.
(c) 010. (d) 101.
(e) 0101. (f) all.
Fig. 11. Upset rate comparison for di®erent values of Vdd.
An Experimental Study of Metastability-Induced Glitching Behavior
1940006-17
J C
IRC
UIT
SY
ST C
OM
P D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by V
IEN
NA
UN
IVE
RSI
TY
OF
TE
CH
NO
LO
GY
on
07/0
2/19
. Re-
use
and
dist
ribu
tion
is s
tric
tly n
ot p
erm
itted
, exc
ept f
or O
pen
Acc
ess
artic
les.
usually implies better dynamic behavior, and these results are in line with reports
from related works.
4.5. Supply voltage dependence of pulsing behavior
For easier comparison between the results of the di®erent measurement runs, Fig. 11
groups the results by observed phenomenon (transition, pulse, etc.) instead of the
supply voltage.
The ¯gure recon¯rms that the resolution constants of the master and slave latches
become better with increasing Vdd (higher slope). The handover point between the
master and the slave is, however, unchanged as the pulses (010 and 101) still start at
the same absolute resolution time. This is as expected, since the timing of the falling
clock edge has not changed.
A more detailed inspection of Fig. 11 reveals that that the (absolute) number of
pulses (010, 101) as well as the number of double pulses (0101) is reduced signi¯-
cantly as Vdd rises. This is not unexpected, since in the above subsection we have
observed improved metastability resolution behavior (smaller �) for rising Vdd. As a
consequence of the lower � , the master latch resolves more events within the given
constant measurement timec leaving fewer events to be resolved by the slave latch.
Therefore, naturally, the number of pulses degrades.
We have computed the relative number of di®erent types of pulses among all
metastable upsets of the slave and found that this yields approximately the same
values all over the supply voltage range. As a consequence, one can conclude that
supply voltage variations do not directly change the pulsing behavior, the in°uence is
only indirect through the changes in � .
cHad we adapted the measurement time to the respective � values, we could expect the same number ofmetastable upsets at the slave latch.
Fig. 12. Number of double pulses.
T. Polzer, F. Huemer & A. Steininger
1940006-18
J C
IRC
UIT
SY
ST C
OM
P D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by V
IEN
NA
UN
IVE
RSI
TY
OF
TE
CH
NO
LO
GY
on
07/0
2/19
. Re-
use
and
dist
ribu
tion
is s
tric
tly n
ot p
erm
itted
, exc
ept f
or O
pen
Acc
ess
artic
les.
To further highlight this e®ect of the supply voltage on the double pulse count,
Fig. 12 shows the number of observed double pulses for both delay lines.d
5. Conclusion
The key aim of this paper was to investigate the pulsing behavior of a °ip-°op that
may be caused by metastability. While late transitions are the more dominant e®ect,
pulses may be more dangerous to the subsequent logic stages, as they may cause
inconsistent behavior, therefore it is important to understand their generation and
the involved dependences.
Since pulsing behavior is much more di±cult to assess by measurement than late
transitions, we had to employ a measurement setup that is substantially improved
over the traditional approach. It is based on using two carefully interleaved time-to-
digital converters in parallel, in conjunction with several calibration steps. This setup
has allowed us to successfully record pulses and establish several ¯ndings:
First of all, we have given clear evidence for the occurrence of pulses in context
with metastability. While the occurrence of single pulses in one direction directly
follows from a simplistic model of the interaction between metastable storage loop
and subsequent decoupling element, pulses in the other direction, as well as double
pulses, were unexpected observations. As a consequence of our analysis we can
summarize the types of metastable behavior as follows: late transitions in both
directions, pulses in both directions and double pulses in one direction (i.e., all with
the same starting edge). We did not observe any pulses of higher multiplicity
throughout our very comprehensive experiments, so we may conclude that the ob-
served pulses are not due to ringing or oscillation.
Furthermore, we have analyzed the dependence of metastable behavior on the
supply voltage. Through that, we could recon¯rm the dependence of � on Vdd that
has been reported in the literature already. When applying our quantitative results
to the observations made in Ref. 28, where three di®erent FPGA boards of the same
type were investigated with very di®erent results concerning their metastable be-
havior, one can conclude that the variations in the � values observed there cannot be
due to supply variations alone. Consequently, there must be a nonnegligible process
variation as well.
Our detailed study of pulsing behavior has yielded a relatively constant propor-
tion of all pulse types all over the supply voltage range. This suggests that pulse
behavior is not directly voltage-dependent, and the only relevant voltage dependence
is in � . As a consequence, the di®erent types of pulse behavior reported in previous
work28 are not likely to be a consequence of supply voltage mismatches among the
di®erent boards. They rather seem to be rooted in process variations through which
the relation between metastable voltage of the latch and threshold voltage of the
dPlease note that the results for delay line B where inverted to compensate for its inverting implemen-tation and are therefore denoted as B�.
An Experimental Study of Metastability-Induced Glitching Behavior
1940006-19
J C
IRC
UIT
SY
ST C
OM
P D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by V
IEN
NA
UN
IVE
RSI
TY
OF
TE
CH
NO
LO
GY
on
07/0
2/19
. Re-
use
and
dist
ribu
tion
is s
tric
tly n
ot p
erm
itted
, exc
ept f
or O
pen
Acc
ess
artic
les.
subsequent bu®er di®ers on the individual FPGAs. This hypothesis, however,
requires further con¯rmation.
More generally, further research is required to ¯nd a conclusive explanation for
the occurrence of double pulses and of single pulses in both directions. Here, an
extensive experimental study must be accompanied by appropriate circuit modeling
that embraces both latches that constitute a °ip-°op, rather than just one.
Acknowledgments
This research was partially supported by the Austrian Science Fund (FWF) Project
SIC (Grant P26436-N30).
References
1. H. J. Veendrick, The behaviour of °ip-°ops used as synchronizers and prediction of theirfailure rate, IEEE J. Solid-State Circuits 15 (1980) 169–176.
2. T. Polzer, F. Huemer and A. Steininger, Re¯ned metastability characterization using atime-to-digital converter, Microelectron. Reliab. 80 (2018) 91–99.
3. D. J. Kinniment, Synchronization and Arbitration in Digital Systems (Wiley, 2007).4. D. Kinniment and D. Edwards, Circuit technology in a large computer system, Proc.
Conf. Computers-Systems and Technology (Institution of Electronic and RadioEngineers, 1972), pp. 441–450.
5. T. J. Chaney and C. E. Molnar, Anomalous behavior of synchronizer and arbiter circuits,IEEE Trans, Comput. C-22 (1973) 421–422.
6. Y. Semiat and R. Ginosar, Timing measurements of synchronization circuits, Proc. IEEENinth Int. Symp. Asynchronous Circuits and Systems (IEEE, 2003), pp. 68–77.
7. D. J. Kinniment, K. Heron and G. Russell, Measuring deep metastability, Proc. IEEE Int.Symp. Asynchronous Circuits and Systems (IEEE, 2006), pp. 2–11.
8. M. Bhushan, M. B. Ketchen and K. K. Das, CMOS latch metastability characterizationat the 65-nm-technology node, Proc. IEEE Int. Conf. Microelectronic Test Structures(IEEE, 2008), pp. 147–151.
9. B. M. Rogina, P. Skoda, K. Skala, I. Michieli, M. Vlah and S. Marijan, Metastabilitytesting at FPGA circuit design using propagation time characterization, Proc. 2010 East-West Design and Test Symp. (IEEE, 2010), pp. 80–85.
10. T. Kacprzak and A. Albicki, Analysis of metastable operation in RS CMOS °ip-°ops,IEEE J. Solid-State Circuits 22 (1987) 57–64.
11. I. W. Jones, S. Yang and M. R. Greenstreet, Synchronizer behavior and analysis, Proc.2009 15th IEEE Int. Symp. Asynchronous Circuits and Systems (IEEE, 2009), pp. 117–126.
12. S. Beer and R. Ginosar, A new 65nm LP metastability measurement test circuit, Proc.2012 IEEE 27th Convention of Electrical and Electronics Engineers in Israel (IEEE,2012), pp. 1–4.
13. S. Yang and M. R. Greenstreet, Simulating improbable events, Proc. 2007 44th ACM/IEEE Design Automation Conf. (IEEE, 2007), pp. 154–157.
14. J. Zhou, D. J. Kinniment, C. E. Dike, G. Russell and A. V. Yakovlev, On-chip mea-surement of deep metastability in synchronizers, IEEE J. Solid-State Circuits 43 (2008)550–557.
T. Polzer, F. Huemer & A. Steininger
1940006-20
J C
IRC
UIT
SY
ST C
OM
P D
ownl
oade
d fr
om w
ww
.wor
ldsc
ient
ific
.com
by V
IEN
NA
UN
IVE
RSI
TY
OF
TE
CH
NO
LO
GY
on
07/0
2/19
. Re-
use
and
dist
ribu
tion
is s
tric
tly n
ot p
erm
itted
, exc
ept f
or O
pen
Acc
ess
artic
les.
15. C. Foley, Characterizing metastability, Proc. Second Int. Symp. Advanced Research inAsynchronous Circuits and Systems (IEEE, 1996), pp. 175–184.
16. P. Alfke, Metastable recovery in Virtex-II Pro FPGAs (2005), Application Note:Virtex-II Pro Family XAPP094, https://www.xilinx.com/support/documentation/application notes/xapp094.pdf.
17. J. Wu, Y. Ma, J. Zhang, Y. Kong, H. Song and X. Han, Research on metastability basedon FPGA, Proc. 2009 9th Int. Conf. Electronic Measurement & Instruments (IEEE,2009), p. 4.
18. J. Kalisz and Z. Jachna, Dual-edge late-transition detector for testing the metastabilitye®ect in °ip-°ops, Proc. 16th Int. Conf. Microelectronics (IEEE, 2004), pp. 780–782.
19. D. Chen, D. Singh, J. Chromczak, D. Lewis, R. Fung, D. Neto and V. Betz, A compre-hensive approach to modeling, characterizing and optimizing for metastability in FPGAs,Proc. 18th Annu. ACM/SIGDA Int. Symp. Field Programmable Gate Arrays (ACM,2010), pp. 167–176.
20. F. Rosenberger and T. J. Chaney, Flip-°op resolving time test circuit, IEEE J. Solid-StateCircuits 17 (1982) 731–738.
21. S. Beer, R. Ginosar, M. Priel, R. Dobkin and A. Kolodny, The devolution of synchroni-zers, Proc. 2010 26th Convention of Electrical and Electronics Engineers in Israel (IEEE,2010), pp. 94–103.
22. T. Polzer, F. Huemer and A. Steininger, A programmable delay line for metastabilitycharacterization in FPGAs, Proc. 2016 Austrochip Workshop on Microelectronics(IEEE, 2016), pp. 51–56.
23. T. Polzer and A. Steininger, An approach for e±cient metastability characterization ofFPGAs through the designer, Proc. 2013 IEEE 19th Int. Symp. Asynchronous Circuitsand Systems (IEEE, 2013), pp. 174–182.
24. R. Najvirt, T. Polzer and A. Steininger, Measuring metastability with free-running clocks,Proc. 2017 23rd IEEE Int. Symp. Asynchronous Circuits and Systems (IEEE, 2017),pp. 18–24.
25. J. Zhou, D. Kinniment, G. Russell and A. Yakovlev, Adapting synchronizers to the e®ectsof on chip variability, Proc. 2008 14th IEEE Int. Symp. Asynchronous Circuits andSystems (IEEE, 2008), pp. 39–47.
26. S. Beer and R. Ginosar, A model for supply voltage and temperature variation e®ects onsynchronizer performance, IEEE Trans. Very Large Scale Integr. (VLSI) Syst. 23 (2015)2461–2472.
27. J. Y. Won and J. S. Lee, Time-to-digital converter using a tuned-delay line evaluated in28-, 40-, and 45-nm FPGAs, IEEE Trans. Instrum. Meas. 65 (2016) 1678–1689.
28. F. Huemer, T. Polzer and A. Steininger, Using a duplex time-to-digital converter formetastability characterization of an FPGA, Proc. 2018 IEEE 21st Int. Symp. Design andDiagnostics of Electronic Circuits & Systems (IEEE, 2018), pp. 141–146.
29. T. Polzer, F. Huemer and A. Steininger, Measuring metastability using a time-to-digitalconverter, Proc. 2017 IEEE 20th Int. Symp. Design and Diagnostics of ElectronicCircuits Systems (IEEE, 2017), pp. 116–121.
30. Xilinx, Zynq-7000 SoC (Z-7007S, Z-7012S, Z-7014S, Z-7010, Z-7015, and Z-7020): DC andAC switching characteristics (2018), Product Speci¯cation, Version 1.20.1, https://www.xilinx.com/Support/documentation/data sheets/ds187-xc7z010-xc7z020-Data-Sheet.pdf.