Text preview for : AR132.pdf part of HP AR132 HP Publikacje AR132.pdf
Back to : AR132.pdf | Home
Comparing the Accuracy and Repeatability of On-Wafer
Calibration Techniques to 110GHz
Anthony J Lord
Cascade Microtech Europe Ltd, 3 Somerville Court, Banbury Business Park, Adderbury, Oxfordshire, UK
[email protected]
Abstract Methods & Limitations
Many methods of making corrected S-Parameters Three different calibration standard substrates were
measurements are available for on-wafer devices and used for the comparisons. One GaAs substrate for the
circuits. This is a comparative study of calibration NIST Multi-Line (LRL) calibration [1], and two alumina
techniques, presented as most accurate and repeatable for substrates for Short-Open-Load-Through (SOLT), Line-
making on-wafer measurements. Reflect Match (LRM) and Line-Reflect-Reflect-Match
(LRRM) calibrations. One alumina substrate being 625um
Introduction thick, and the other 250um thick. As a recommendation
from Ref. [2], the thin ISS included a layer of Radiation
An on going concern when making on-wafer Absorption Material (RAM) between the Impedance
calibrations and measurements is exactly how accurate and Standard Substrate and metal chuck surface.
repeatable are the measurements you're making. Because A major limitation of the paper is lack of a reliable
of the complexity and diversity of the measurement system precision reference measurement, to 110GHz. An
it makes traceability back to a physical reference extrapolation was made from the results of Ref. [3] to
impractical. We can however compare the complete cover the higher frequency band. The NIST LRL
measurement system, including probes, calibration calibration standards are not a modelled 50ohm
standards and algorhythms to a benchmark standard transmission line to 110GHz and a miss-match to 50ohm
defined by the National Institute of Standards and calibrations can be expected. My LRL calibration
Technology (NIST). With the growing interest in reference planes were at the centre of the 500um thru' line,
millimeter-wave devices due to growth in the aerospace, and the Zo was referenced to the Line. To compare the
automotive and optical industry, it is important to common calibration methods used by engineers today for
understand which calibration set up will offer the most on-wafer microwave measurements I have performed
superior measurement performance for any particular several calibrations using SOLT, LRM, LRRM with Auto
application. Load Inductance Compensation [4], and LRL.
0.5
Measurements were collected, using each resulting
0.4
calibration co-efficients, of both active and passive devices
to determine if a measurement difference is apparent by
0.3 LRM
using different techniques. A commercially available
0.2
software package [5] was used for performing calibrations
0.1 and recording measurements.
dB
0.0
-0.1 Measurements & Results
-0.2
SOLT
-0.3
Open Circuit Measurement
0.5
-0.4
0.4
-0.5 Standard ISS Cals
0 10 20 30 40 50 60 70 80 90 100 110 0.3
[GHz]
0.2
Fig. 1. Measurement of open standard after calibration 0.1
falsely shows SOLT to be perfect, which is a result of the dB
0.0
SOLT calibration forcing the reflections to be 0.0dB
-0.1
-0.2
To identify the true integrity of the SOLT calibration
we require independent verification standards. Re- -0.3
Thin ISS w/RAM
measuring the same standards will only show the -0.4
repeatability of the -0.5
Contact. This is shown in Fig. 1. The SOLT calibration is 0 10 20 30 40 50 60
[GHz]
70 80 90 100 110
not self-consistent and the open circuit response shows a Fig. 2. Measurement of open standard after SOLT, LRM
perfect reflection, where the LRM calibration method is & LRRM calibrations.
self-consistent and errors can be identified looking at the
magnitude of Sii. It is not a safe assumption to believe The open standard measurements using the SOLT
SOLT is more accurate because it looks like a perfect calibrations co-efficients indicates a near perfect reflect,
open. since we are only performing a repeatability measurement
of the contact. The thinned 250um ISS and layer of RAM 0.0
material reduced the magnitude of error on both LRM and
LRL Cal
LRRM calibrations. The large error using the 625um thick -0.5
ISS was due to the substrate moding being more significant
at millimetre wave frequencies. The 250um ISS pushes -1.0
the substrate moding above 110GHz. This now meets the
commonly used error limits of +/-0.1dB for open circuit dB
-1.5
verification.
-2.0
ISS Cals
Open Stub Measurement
A more reliable way of verifying the integrity of the -2.5
calibration is to measure an independent verification
standard. I used a 3.2mm open stub and 3.2mm line of the -3.0
0 10 20 30 40 50 60 70 80 90 100 110
NIST reference substrate. The ISS calibrations (LRM, [GHz]
LRRM and SOLT), using both the 625um and 250um thick Fig. 5. S21 LogMag measurement of NIST 3.2mm line
substrates, show a ripple effect. This is due to the line not
being exactly 50ohm and is miss-matched to our 50ohm S21 LogMag Variation from LRRM/250um ISS
ISS calibrations. The LRL calibration shows a more linear
response, but a phase and magnitude offset is present due 0.3
to the reference plane being in the centre of the LRL thru,
not the probe tips, as with the ISS calibrations. 0.2
SOLT
0.0
-0.5 NIST LRL 0.1
LogMag (dB)
-1.0
-1.5 0
-2.0
dB -0.1
-2.5
-3.0
-3.5 ISS Cals -0.2
-4.0
625um ISS
-4.5 -0.3
Freq 0-110GHz
-5.0
0 10 20 30 40 50 60 70 80 90 100 110 Fig 6. LogMag variations of line using LRRM/250um ISS
[GHz]
as reference. (ISS calibrations only)
Fig. 3. S11 LogMag measurement of 3.2mm open stub.
S 2 1 P h a s e V a r i a t i o n f r o m L R R M / 2 5 0 u m I SS
55 Long Open Verification 1.5
0
-0.5 1
-1 0.5
S11 ref Z0=55
-1.5
Phase (deg)
0
-2
-2.5 -0.5
-3 -1
-3.5
-1.5
-4 S11 ref Z0=50
-2
-4.5 625um ISS
-5 -2.5
0 10 20 30 40 50 60 70 80 90 100 110
Freq 0-110GHz
Frequency (GHz)
Fig. 7. Phase variation of line using LRRM/250um ISS as
reference. (ISS calibrations only)
Fig. 4. Model of 3.2mm open verification standard,
making the assumption that the GaAs line is not 50ohms The ISS calibrations have approximately the same
deviation from the LRL measurement, as shown in Fig. 5.
Line Measurement Using the LRRM calibration as a reference, the variation
The GaAs line measurement show the LRL being of the LRM and SOLT calibrations can be observed. The
comparable to the ISS based calibrations up to 70GHz, 625um ISS and SOLT calibrations show greater variation
where afterwards the ISS calibrations shows greater loss. in phase and magnitude. The phase variation of the ISS
This may be a result of the miss-matched line acting as a calibrations from the LRL calibration shows a linear phase
low pass filter for the 50ohm calibrations.
change due to the reference planes of my LRL calibration
being the centre of the 500um thru' standard and not the tip S21 Phase Variation from LRRM/250um ISS
of the probes as with the ISS calibrations. 5
SOLT/250um
4
Field Effect Transistor (FET) Measurement ISS
The measurement accuracy very much relies on the 3
calibration and the measurement application. Fig. 8 shows 2
a measurement made of a GaAs FET device. The SOLT,
Phase (deg)
1
LRM and LRRM calibrations are grouped together. The
only stray measurements are the NIST LRL calibration. 0
The difference between the LRL and other calibrations is -1
probably not due to inaccuracy of the ISS based
-2
calibrations. It is likely due to the inaccuracy of the LRL 625um ISS
calibration due to the change in pad parasitic, the change of -3
effective dielectric constants and the low-end limitation of -4
the calibration due to the restrictions of long line standards. Freq 0-110GHz
Fig. 10. S21 Phase variation of GaAs FET device with
reference to LRRM calibration using 250um ISS.
15
Repeatability of Calibrations
NIST LRL The need to make an accurate calibration and
10 measurement is equalled by the requirement to make
repeatable calibrations and measurements. It is shown in
Fig. 11 and 12 the worst case error bounds for repeating
dB
5 two identical calibration techniques. The results show that
the LRRM calibration with load inductance compensation
was more repeatable than SOLT, which was particularly
0
sensitive when using different sets of standards.
ISS Cals
-10
-5
0 10 20 30 40 50 60 70 80 90 100 110
-20 LRRM with load inductance
[GHz]
compensation
Fig. 8. Measurements of a GaAs FET device. SOLT
-30
The SOLT calibration performed on the 250um ISS dB
-40
indicates a linear increase in magnitude and phase, Fig. 9
& 10. The SOLT, LRM and LRRM calibrations -50
performed on the 625um ISS shows the same errors when
measuring the open circuit during calibration verification. -60
Only the LRM calibration made on the 250um ISS is
comparable to the LRRM reference calibration. -70
0 10 20 30 40 50 60 70 80 90 100 110
[GHz]
S21 LogMag Variation from LRRM/250um ISS Fig. 11. The worst case errors for calibration repeatability
using the same set of standards
0.6
-10
0.5
SOLT
0.4 -20
SOLT
0.3
LogMag (dB)
-30
0.2
0.1
dB
-40
0
-0.1 -50
-0.2
LRRM with load inductance
-60
compensation
-0.3
-0.4 625um ISS
-70
Freq 0-110GHz
0 10 20 30 40 50 60 70 80 90 100 110
[GHz]
Fig. 9. S21 LogMag variation of GaAs FET device with Fig. 12. The worst case errors for calibration repeatability,
reference to LRRM calibration using 250um ISS. using two different sets of standards.
I performed eight LRRM calibrations using the same The FET device results identified large variations at
set of ISS standards, but replacing the probes manually on low and high frequencies between the LRL calibration and
the ISS alignment mark. Even though my probe placement the ISS based calibrations. The low-end variation was a
was not exact due to the limitation of the optics and limitation due to the line length required for low
resolution of the positioners, the open standard verification frequencies and the large imaginary component of the
has a worst case spread of 0.15dB. The same experiment characteristic impedance at low frequencies due to
was repeated but using eight different sets of standards. conductor resistance. The high frequency was a result of
The repeatability of calibration was decreased, but only differences in pad parasitic between the calibration
marginally, to 0.2dB. All the calibration verifications were standard and DUT.
within the general recommended limits of +/- 0.1dB up to The 625um thick ISS exhibited a larger error in
110GHz. Open measurements phase error is expected to be magnitude when verifying the calibration, using an open
more sensitive to probe placement errors causing small standard. This error is noticeable when measuring a
changes in reference plane location. reflective DUT such as an open or open stub and was also
0.20
noticeable on the S21 of a FET measurement.
Whilst performing the calibrations, my observations
0.15
included how essential probe placement accuracy was for
0.10 all calibration methods, but was even more so important
when making LRL and SOLT calibrations. The probe
0.05
placement error was not critical when using load
dB
0.00 inductance compensation, which was used for the LRRM
measurements. Several calibration attempts were required
-0.05
to achieve satisfactory results for the techniques not using
-0.10 0.15dB Worst case error load inductance compensation. Indeed, I encounter long
-0.15
and tedious problems trying to achieve a `good' NIST LRL
calibration, and it was not easy to achieve repeatability.
-0.20 Also whilst making my calibrations it was noted that a
0 10 20 30 40 50 60 70 80 90 100 110
[GHz] good LRRM calibration with load inductance
Fig. 13. Sii Open measurement of 8 LRRM calibrations compensation was achieved after every attempt. The
repeatability of making numerous LRRM calibrations
0.20
proved to be better than