Article Reprint
RF (Gigahertz) ATE Production Testing
On-Wafer: Options and Tradeoffs
Reprinted with permission of IEEE.
Copyright ©1999 by the Institute of Electrical and Electronic Engineers.
This paper was presented by Dean Gahagan of Cascade Microtech at the
1999 International Test Conference.
This paper describes the hardware solution tradeoffs in testing RF devices on-wafer in
a production environment using ATE. The options, which are available today, will be
compared with respect to RF measurement integrity and production worthiness.
Introduction
The ever-increasing demand for more bandwidth to support the exploding
Internet and the insatiable need for PCS (personal communication
systems)/cellular devices has driven manufacturers of high speed digital and
RF devices to push RF test of these devices upstream to the wafer level for a
variety of reasons. Internet user demand for bandwidth is doubling every four
months, while providers are only able to double bandwidth access every fourteen months. Worldwide sales of cellular telephones were in the hundreds of
millions in 1998. The high volume demand of these devices necessitates the
most economical production path to market.
Why RF test on-wafer?
The relentless push for IC cost reduction has quickly driven cost-sensitive RFIC
integration levels from single-function parts in 1995 to full front ends and simple
handset functions on a chip today [ref 1]. Small RFICs are usually packaged with
little or no RF test (sometimes not even a DC sort), and typically packaged in
cheap SOIC packages. The only RF testing is done at final test, since package scrap
costs are very low, although sample testing for RF functionality before assembly is
useful to ensure that there are no bad lots in the packaging queue. As the IC complexities have increased, the yields are lower and the package costs are higher, creating a need for screening before packaging to save package scrap. In the future, integration levels will continue to rise, and package inductance requirements will imply
chip-scale packages (CSPs) and/or flip-chip assembly with known-good die (KGD).
Figure 1 illustrates this evolution.
1995: SSI parts in SOP’s
PCM
Test
Sample RF Test
At Wafer Level
Dice &
Package
100% Final Test
In Package
1998: MSI parts in plastic or BGA
PCM
Test
100% RF Test At
Wafer Level
Dice &
Package
2001: MSI or LSI as CSP or KGD
PCM
Test
100% RF Test At
Wafer Level
Singulate
Die
Fig. 1. Current and predicted test flows for RFICs.
100% RF Or Connectivity
Test In Package
High-performance RFICs, such as military/aerospace or SONET applications,
are typically packaged in more expensive packages or modules, necessitating RF
screening at the wafer level for KGD. Virtually no RFICs are burned in today.
In the past 15 years, wafer probing above one GHz has progressed from impossible, to a useful R&D tool, to a necessary production tool, to a mainstream VLSI
topic. In the early 1980’s, it was common knowledge that you could not get good
measurements from a wafer probe at high frequencies. In the late 80’s, the GaAs
MMIC community had adopted the first ceramic-based microwave probes as a necessary tool for device and IC development. By the early 90’s, production microwave
and high-speed IC’s bound for expensive modules or packages were being 100% RF
probed before assembly. In the late 90’s, mainstream CMOS now routinely achieves
30 GHz fT, allowing large IC’s that operate at 500 MHz clock rates and higher.
These high-value chips are often packaged in expensive multi-chip modules
(MCM’s), requiring known-good-die (KGD) screening in production at microwave
bandwidths on tens, and soon hundreds, of signal and power contacts.
Required
Measurements
The RF and IF measurements required, test times, number of RF channels supported and bandwidth of measurements are all parameters that influence a test
engineers hardware choices for ATE. Figure 2 is a list of the measurements that a
RFIC designer may wish to perform in verifying or characterizing an IC. A different subset of these would typically be performed, depending upon the type of
device under test (DUT), e.g., passive filters, amplifiers, mixers or power amplifiers.
Classically these measurements have been done with rack-and-stack RF test gear.
Fig. 2. Typical RF and IF measurements made on a RF device
AM-PM conversion (static)
Adjacent channel power
Complex demodulation
Digital input-threshold voltage
Digital output levels
Efficiency (RF out/DC in)
Frequency
• cw
• vs time (tune drift)
Gain or loss
• vs control voltage or digital state
Gain compression
• Pout @ N dB, saturation
Harmonic distortion
• dBc
• SOI, TOI
I/Q modulator imbalance (static)
• amplitude & phase error
I/Q modulator suppression
• carrier & unwanted sideband
Isolation
Minimum detectable signal
Mixer conversion gain or loss
Mixer leakages
• LO-->RF, LO-->IF, carrier feedthrough
Noise figure
Nth order intermodulation
• two-tone IP3, IP5
Phase noise/jitter (cw)
• modulated
Power (dBm)
• output power
• vs bias voltage
Pulsed RF measurements
• frequency
• power
• S-parameters
Pulsed RF profile signal overshoot & ringing
RF rise time (10% to 90%)
S-parameters
• gain/loss isolation, match, VSWR, gamma
Spurious signals
• @ known frequency, search
Supply currents
• enabled, sleep mode
Switching speed
• D digital input to D gain or D frequency
VCO frequency
• vs voltage
• tune linearity
• tune range
• tune sensitivity (dc-freq)
• vs digital state
Voltages
VSWR
HP 84000 RF test configuration
Tester to Probe Card
Interfaces
Recently, RF ATE systems have become readily available that integrate all of these
functions and provide a much higher throughput capability than typical rack-andstack equipment. These RF ATE systems with various options are available from
numerous companies. Agilent Technologies, Teradyne, LTX, Roos, Advantest,
TMT, and Credence all offer giga-hertz testers. The details of these measurements
are beyond the scope of this paper, but all of the ATE vendors will provide application and benchmarking information for these measurements. The major trade-offs
one needs to be aware of are number and quality of RF ports offered, number of
synthesized RF sources, the amount of RF output power, and number of DC and
AC channels. Types of measurements the system can do and the software interface
are critical needs the production test engineer has to understand. The system architecture is pivotal to the level of calibration one can do on the system and the interfacing components that mate the ATE to the DUT. Most take a tiered approach
where the first tier is the test head (sometimes referred to as the docking head) and
subsequent tiers are for the various pieces of interface hardware.
Usually three options can be used to interface the test head to the probe card.
1) Cabling from the test head to the probe card is an option where there is no
requirement of a docking or test head that has to be positioned on and off the
probe station.
2) A probe interface board (PIB or sometimes referred to as a tester load board)
with a mechanical fixture that ties the load board and the probe card together is
another way of interfacing to the test head.
3) The last option is a probe board that mates directly to the test head.
Measurement requirements, easy configurations, production worthiness, and cost
are trade offs when you make this choice. Cabled interface, tower fixture, and a
direct dock interface will be the terms that will be referred to in this paper.
Cabled Interface
Major advantages of the cabled interface are flexibility in configuring your tester for
many different types of devices and typically lower cost. Cost saving occurs from
not having to carry an inventory of expensive load boards, pogo-pin towers, and
other hardware in stock to test different applications or devices. There is also no
requirement for a manipulator to move the test head on and off the probe station.
Major disadvantages are the complexity of changing from one device to another
in a production environment. There is an additional cost here of having a skilled
person available on the production floor that can change all of the complex connections and then do some type of calibration or verification that the system is
operating properly and making the correct measurements.
From an RF measurement standpoint there is usually some cable loss associated
with this type of interface and there is a real risk of having intermittent connections
at the connectors if they fail or are not tightened properly. The test engineer has to
make the decision whether the RF loss through the cables can be removed through
calibration or whether the output power from the tester or the DUT is sufficient to
overcome all of the losses through the cable and probe card.
Tower interface
Tower interface with theta
adjustment
Tower Fixture
The tower fixture has a load board that usually has blind-mate RF connectors,
which provide the first interconnect of a RF signal to or from the test head. Blindmate connectors can tolerate some mis-alignment tolerance of the tower fixture to
the test head and still provide a good RF connection. DC and AC connections to
the test head are accomplished by either a pogo-pin configuration or some type of
square pin connection arrangement. Some type of mechanical guide and locking
mechanism ensures proper mechanical alignment and mating of this fixture to the
test head. A theta adjust is also provided on some systems.
Board real estate is the major advantage of using a load board instead of just
direct docking the probe card to the test head. Any custom circuitry that is critical
to being close to the test head but not to the DUT can be placed on the tester load
board, which can be application or device specific. Components can consist of
relays, bias tees, transformers, test head terminations, filters, amplifiers, or any
other passive or active component the test of the device requires. Semi-rigid or
flexible coax can be used to route RF lines from the load board to the probe board.
Semi-rigid provides a loss advantage as well as mechanical and thermal stability.
Flex cable is an easier routing solution. The shorter the tower, the less loss in
the cables. Tower configurations are usually the most costly as you have a load
board, the mechanical and electrical configuration between the two boards,
and the probe card. Once this tower is assembled for a given DUT, it usually
remains together until some repair is required. Spare towers are a costly but necessary inventory in a high volume production environment. Manipulator arms are
usually required in the tower setup to lift the test head and tower interface off the
probe station.
The real benefit of a tower interface is it allows you almost the same flexibility
as the cabled interface, yet provides an assembly that is very easy to set up, reliable
in a high volume production environment, and maintains good RF performance.
RF losses are minimized, and insertion loss and delay can be calibrated out in the
frequency domain, moving the reference plane of the measurement out to the
probe tips. Individual wires, ribbon cable, or even coax (to reduce noise on these
lines) are accept ways of connecting the DC and AC signals from one end of the
tower to the other.
Direct Docking
Interfacing the probe card directly to the test head has some distinct advantages.
The probe card has blind-mate RF connectors that mate directly to the test head.
RF losses are minimized and the ability to take a probe card off the shelf and easily
attach it to the test head makes this solution the most beneficial in a production
environment. An inventory of probe cards can be maintained with a minimum of
cost where there is no elaborate mechanical fixtures and no expensive and complex
load boards. The disadvantage of this type of interface is the only place supporting
components for the IC or circuitry needed to customize the test or test head is on
the probe card. There is also no mechanical isolation between the test head and the
probe station.
Probe Cards
The performance of the probe card is sometimes the least understood section of the
entire measurement system. Hundreds of man hours of engineering and hundreds
of thousands (if not millions) of dollars are spent on the tester design, the test
head, the load board and tower interface, the manipulator, and the software to execute a test sequence. A minimal amount of effort is usually spent on the probe card
design and is usually saved until last when the wafer is in fabrication and the die
pads are fixed. With all of the careful RF design considerations taken into account
with the rest of the measurement system, the probe card is truly where the rubber
meets the road. A great deal of effort is spent in controlling parasitics, bypassing,
and controlling impedances in designing a package so this device works to its maximum performance. This consideration should also be taken when looking at the
design of a probe card. If this care is not taken, all of your measurement quality
paid for in the tester could be lost as all I/O signals from the DUT pass through
the probe card.
Quadrant coaxial and blade probes
Parameter
Epoxy-ring
Needle Card
Coax plus
Needles Card
RFIC PyramidTM
Probe Card
Maximum pad count
500
80
800
Parallel test?
No
No
Yes
RF line bandwidth
1 GHz (best case)
110 GHz
110 GHz
10 – 20 nH
3 – 8 nH
0.4 – 0.2 nH
< -1 dB
-1 to –4 dB
-20 to –26 dB
Common-lead
• Inductance
• Single-ground
crosstalk @ 2 GHz
Table 1 – Performance comparison of different probe card technologies
The card has blind-mate RF connectors
that mate directly to the test head
Table 1 compares the available technologies for RFIC probe cards. Any of these
technologies achieve losses (attenuation) that are low enough to be corrected in any
RF measurement. The problem becomes the other fixturing parasitics. The return
loss of a typical needle card is not suitable at 2 GHz, since this does not achieve an
uncorrected return loss for the whole system of at least 10 dB. The coax plus needles or a pyramid card easily achieves the return loss corrections. The line-to-line
crosstalk of a needle card is also unacceptable since the crosstalk is not correctable
in most microwave measurements. The coax plus needle cards achieves good coupled-line crosstalk, but there is an RF signal density constraint of at least 500 µm
just to fit the coax lines near the probe tips.
The RFIC probe cards achieve by far the best coupled-line crosstalk. The common lead inductance is much too high in a typical needle probe, either for the
ground or powers, and it is also way too high in a coax plus needles card arrangement. Only the RFIC Pyramid probe card achieves the inductance control for
accurate emulation of package parasitics. [Ref 2] Power bypassing on a probe card
is crucial in delivering clean power and low ground inductance to the DUT. Needle
and coaxial probe cards do not allow bypass capacitors to be placed close enough to
the DUT and when one tries to put them close, there is still a considerable amount
of lead inductance between the device and the bypass capacitor. The membrane
probe allows low impedance microstrip lines to connect bypass capacitors between
power and ground. The ground inductance on the membrane card is sometimes an
order of magnitude or less than other types of probe cards. Measurement repeatability and correlation are essential in high volume production testing.
Bottom of Pyramid Probe card
Calibration
Photolithiographical probe cards are inherently the same from probe card to
probe card. RFIC probe card options are limited to blade needle cards with coax
probe blades, or membrane-style probes. Coaxial blade cards are able to contact
three or four widely-spaced single-ended RF ports through 110 GHz, but have
poor ground and power bypassing parasitics. Above about 1 GHz, membrane-style
probes are the only option offering high density, low power and ground impedances, or element integration close to the IC pads.
Most of the major ATE systems for RF production testing have some capability of
performing calibration corrections to the measurements at the test head, the fixture
interface, and the probe card. Most use a tiered approach where a test head calibration board is inserted into the test head to calibrate the ATE. A standard for the
interface can then be used to calibrate out standard test interface fixtures or probe
cards. Calibration standards on-wafer or on a calibration substrate can then be used
to calibrate down to the probe tips. A short, an open, a load and a thru are then
required to perform a full VNA calibration. There are different algorithms (LRM,
LRRM, SOLT, SOLR) used to calculate the calibration coefficients. The ATE systems either have scalar or vector analysis capability.
Calibration plane
(@ precision blindmate
connectors)
Measurement
calibration with
calibration
boards
ED
Measurement plane
(@ probe tips)
De-embed
•Loss
•Phase shift
•Mismatch
ES
ET
Calibration plane extended to probe tips with de-embedding
General purpose membrane impedance
standard substrate
Scalar network analyzer (SNA) or power meter setup normalized for cable
and/or fixture losses will correct for the losses but be in error to the extent that the
DUT reflections are different from the calibration through reflections. A VNA will
correct for both losses and reflections, but only if the DUT is linear. (A rule of
thumb is that the VNA can improve the uncorrected measurement system directivity by about 20 dB.) Most importantly, there are no practical general methods of
correcting any microwave measurements for crosstalk of any type. If forced, you
can simulate the IC with different inductances or couplings, but the simplest
approach is to use probe cards and fixturing with low crosstalk.
The effects of series inductances of the input or output leads can be added or
removed from measurements relatively easily, but the effects of common-lead
impedances are very difficult to correct even in the easiest cases. The best solution
is to carefully control the actual probe card and fixture inductances and power
supply impedances.
Probe Stations
EG 4090 autoprober
Summary/Conclusions
Probe stations are another key piece of hardware in making good quality and
repeatable RF measurements in a production environment. Automated wafer
handling, calibration functions, testing devices at temperature, low noise environments, automatic probe to pad alignment, and software integration to the ATE
test executive are all functions needed to perform different RF measurements
discussed above.
ElectroGlas, TSK, TEL, and Cascade Microtech manufacture fully automated
production probe stations capable of performing these functions. Thinned GaAs
wafers require careful handling by automatic probe stations. Robotic arms are
capable of this task. Auxiliary chucks should be available to place calibration substrates near the chuck so calibration can be performed on a regular basis.
Enclosing the wafer in chamber allows a controlled temperature environment and
minimizes external noise for critical RF measurements such as noise figure.
Calibration software integrated with the probe station can be beneficial to ensure
critical calibration coefficients are measured properly and downloaded to the ATE.
A good understanding of required measurements and knowledge of options for the
total solution are required to make essential tradeoffs between quality of measurements, production worthiness and cost in making good decisions for on-wafer RF
production test. The optimal solution is device specific in nature and the final configuration should take into consideration the applications that a manufacturer
needs to perform on those devices. Multiple vendors are available for most pieces of
the entire system. Some ATE vendors support integrated cells and take responsibility for the total solution. RF wafer test is emerging and production-ready tools are
generally becoming available.
Literature references
[Ref 1] E. Strid, Roadmapping RFIC Test, 1998 GaAs IC Symposium Technical
Digest, pp. 3-6.
[Ref 2] E. Strid, Trends in RF Devices and ICs for Wireless Applications, IEDM Short
Course, p. 41
Texas Instrument, Michael Janssen, Microwave Probetest Evaluation on Teradyne
A585
HP84000 Series, RFIC Test Systems, Randy McBride, A.02.00 RF Wafer Test
Capability
Contribution is to describe practical HW options for on wafer production
RF test and to compare these options with respect to RF Performance and
Production Worthiness.
© 1999 IEEE. Reprinted, with permission, from an article
presented by Dean Gahagan of Cascade Microtech at the
International Test Conference in 1999.
AR134-0701
Data subject to change
without notice