Incredibly, the number of transistors in integrated circuits (ICs) has tracked Moore’s law, doubling every two years. Equally amazing is the recent jump in IC device operating frequency into the mmWave spectrum. The emergence of 5G, next-generation WiFi protocols, and automotive radar have pushed ICs into these extreme frequency bands to take advantage of the additional available bandwidth. The need for instantaneous data transfer has driven the increasing demand for these new mmWave devices, whether for safety concerns in automotive radar, enormous data networks for 5G cellphone backhaul, or simply the expectation to instantaneously stream 4K video from your tablet to your flat screen. Historically, 6 GHz was the high end of the frequency band for the majority of ICs. Operating frequencies have jumped to 30, 40, 60, and 80 GHz to get the necessary bandwidth for next-generation 5G, WiFi, and automotive devices, respectively. This equates not to a doubling in frequency, but to a gigantic leap that in some cases is greater than an order of magnitude. An additional challenge beyond the rapid growth in operating frequencies is caused by 5G applications driving an emerging need for production over the air (OTA) test of antenna in package (AiP) devices.
Today, commercially-available ICs exist that require bandwidth in the GHz range for early 5G applications in various mmWave frequency bands from 30 GHz to nearly 100 GHz. However, for years many of these devices were only engineering samples confined to a characterization lab for in-depth analysis and debug. Recently, mmWave devices, initially from the automotive industry, are moving quickly into the production environments where high-volume testing is required. The quick transition from concept to production has created a bottleneck as device test plans and interface hardware were previously not yet fully defined. The high-volume production test cells used for 6 GHz devices were modified to up-convert, mix, down-convert, and measure mmWave frequencies. This resulted in a very complex and custom interface for each mmWave application.
Interface hardware has now become the critical path for mmWave IC testing. What historically was considered a high-speed load board and spring pin socket no longer provides sufficient performance for this new generation of ICs. Now, the interface must be considered holistically, including all the mechanics of connecting the IC to the tester/handler and the impact the environment has on the transfer of data at incredible speeds. In the mmWave frequency bands, miniscule changes in the environment wreak havoc on the electrical performance. Stack-up tolerances and temperature fluctuations in handler kits, docking hardware, connectors, cabling, etc., all impact electrical performance and the ability to get 100% test coverage of mmWave devices.
Early on the expectation was to bring mmWave devices to production with the assistance of either loopback on the load board or built-in self test (BIST) in the device. Because tester resources for mmWave frequencies were unavailable, test engineers included loopback traces on the load board from TX to RX, or IC designers included mixing and couplers at the die level to sample the mmWave signals without interface to the outside world. Although these methods circumvent the need for a mmWave test system, it either becomes too tedious to debug or consumes valuable real estate and adds complexity to the die, thereby delaying time to market and/or pushing the IC cost upwards. Furthermore, the accuracy and reliability of the test results are not completely understood.
Testers now have first-generation up-conversion, mixing, and down-conversion add-on modules that allow a more traditional automatic test equipment (ATE) test plan. These bolt-on modules extend the frequency capability of the 6 GHz tester to the mmWave region. The typical maximum bandwidth of these add-on modules is around 10 GHz so they are considered banded solutions. Interface hardware can be built to be broadband, but components in the mmWave region are typically band limited, which drives custom hardware for different frequency bands. A 5G module designed for 28-39 GHz, a WiFi module designed for 56 GHz to 64 GHz, or an automotive radar module designed for 76-81 GHz all require a different add-on module for the tester. Besides being banded, these solutions are typically scalar rather than vector to keep costs down. They can measure gain or output power, but not phase. The full vector solution (a vector network analyzer (VNA) in a tester) systems exist, but the increasing radio frequency (RF) channel count of the new devices are making full vector systems prohibitively expensive.
Although most current generation testers now have instrument options that can supply at-speed signals to the device under test (DUT), their calibration plane ends with a power meter at the test head. That means the losses associated with any hardware between the test head and the DUT must be understood, limited and calibrated out. The challenge of mmWave production test is getting sufficient electrical performance at the DUT while maintaining a reliable contact in an environment with significant and unknown stack-up tolerances. When you add the stack-ups of the test environment (docking hardware, stiffener frames, PCBs, contactors, change kits, etc.) it becomes obvious why a compliant interconnect mechanism is required. To maintain reliable contact to the DUT, the interface must include a compliant transmission line without adding significant additional loss to the system.
Traditional ATE interface hardware for 6 GHz applications had enough performance overhead that a standard spring pin and high-frequency load board PCB dielectric was sufficient. Regardless of how these components were put together, the parts could be tested with reasonable yield, however, at mmWave frequencies this is no longer the case. Every component and transition must be optimized for minimal loss and best impedance match. At mmWave frequencies, testers can source +5 to +10 dBm output power and must put the DUTs into saturation for 100% test coverage. DUTs in the mmWave range require approximately -5 dBm input power so a 10 dB loss through the system is the threshold. Furthermore, it becomes difficult to calibrate with higher than 10 dB loss in the interface. The losses mask the difference between calibration standards and limit the accuracy of the calibration.
Loss adds up quickly at 80 GHz. Transitions from one component to the next are a big factor. Even linear loss from traditional load board PCB architecture can quickly overcome the 10 dB threshold. High frequency dielectrics have up to a 3 dB per inch loss at 80 GHz. Traditional spring pin socket loss above 40 GHz quickly degrades to higher than 10 dB and resonances change from insertion to insertion making an unreliable interconnect. Finally, SMA connectors are mode limited to around 27 GHz so cable sourcing may need to change.
At mmWave frequencies, an electrically transparent interconnect is difficult to achieve, but with some creative design practices it is possible to get sufficient electrical and mechanical performance to test beyond 100 GHz. Thinking outside the box of the traditional PCB and socket structure is required. Minimizing the number of impedance discontinuities is a priority. Bypassing the PCB and directly interfacing with the compliant interconnect eliminates the launch into and out of the PCB. Although the PCB is a great interface for high-density low-frequency and power signals, the dense environment extends line lengths and limits optimization at mmWave frequencies. Furthermore, the standard spring pin socket is not a reliable option at mmWave frequencies. For the high-speed signals, a single-piece compliant interconnect, such as Cohu’s xWave technology, is a better option as it eliminates resonances created by the length of the pogo pin and inconsistencies on account of the changing interaction between the probe components. Combining the socket and PCB into one single compliant transmission line provides a low-loss, well-matched interface from the tester to the DUT with minimal discontinuities.
The 5G revolution also includes AiP devices that require OTA test solutions at these mmWave frequencies. OTA test setups are complicated further by application variables, such as near field/far field, antenna integration level, and radiation orientation. The position of the test interface antennae from the device in the test setup could vary between a few millimeters to tens of centimeters depending on whether it is near/far field and the center frequency. These antennae can be integrated at the die, package, or module level. The radiation direction of the antennae can also be oriented to come out of the ball/pad, lid, or side of the device or module. These variables drive the test cell configuration beyond the norm and blur boundaries between the tester, contactor/probe head, and handler/prober to meet the test requirements.
mmWave test interface solutions
Package and wafer test interface solutions exist today for mmWave devices. The front runners of these solutions are membrane, conductive elastomer, and hybrid lead frame based. Because mmWave devices are just now reaching production volumes, the mmWave interface solutions are only now being exercised in production environments. In-depth comparisons on longevity, maintainability and performance are being done today. Each available solution has advantages and disadvantages.
Conductive elastomer solution. Conductive elastomers minimize the inductance of the socket with extremely short interconnects, but they still require a PCB interface for the mmWave signals. They are typically combined with small daughter cards to provide sufficient electrical performance for mmWave frequencies. Elastomers have limited compliance, a relatively short lifetime (typically less than a couple hundred thousand test cycles), and their characteristics vary widely with temperature. These limitations inhibit the technology from being used in production environments where temperature fluctuations and stack-up tolerances can stress the elastomer.
Membrane solution. A membrane solution includes an impedance-controlled flex circuit. The mmWave signals bypass the PCB and transition from the flex circuit into a semirigid coaxial cable and/or an RF connector. The low frequency and power signals transition from the flex circuit into a traditional PCB. Membrane solutions create an impedance-controlled path from the test equipment to the DUT, but they are fragile and have limited compliance, which can lead to reduced lifetime. Membrane solutions are often used in wafer test applications where tolerance stack-ups are reduced and precise prober automation reduces the wear and tear on the interface.
Hybrid solutions. When considering the harsh production environment of test, a lead frame and spring pin hybrid solution, such as Cohu’s xWave technology, offers electrical and mechanical advantages. The xWave contactor or probe head utilizes patented hybrid contacting technology to optimize RF performance and provide robustness for production testing of the most challenging cmWave and mmWave devices. An impedance-controlled lead frame connects the mmWave signals while standard spring pins connect low-frequency and power connections. The hybrid solution embeds coplanar waveguide lead frame structures in the socket with coaxial waveguide connections to the test equipment. The traditional spring pin socket and PCB remains for the low frequency, control and power signals. This approach combines the robustness of the traditional spring pin/PCB interface while eliminating the discontinuities for the mmWave signals. The result is an interface solution that can withstand tri-temp high-volume production environments and last greater than 1.5 million cycles on a handler or greater than 3 million cycles on a wafer prober. The hybrid nature of the xWave technology has also enabled the integration of patch or horn antennas into the contactor either above, below, or beside the DUT depending on the radiation direction(s). Cohu’s entire test cell domain knowledge also allows for the seamless integration of these solutions into the tester or handler as needed to provide optimized solutions.
Regardless of the interface technology applied, a system-level approach to interface design is required. The socket/contactor cannot be considered alone. Interactions with the PCB, probe card, connectors, cables, waveguides, etc., have an impact on performance that cannot be ignored. Applying rule of thumb RF design practices is insufficient when designing mmWave interfaces at 80+ GHz. Complex interactions create unforeseen resonances that cannot be predicted without 3D electromagnetic simulation. The best approach is simulation of the entire path from the tester to the DUT to confirm electrical performance before fabrication. Mechanical tolerances must also be considered. Building worst-case tolerances into the simulation models provides the most conservative result and provides the best probability of first-pass success. Electrical and mechanical engineering co-design is required and needs to consider mechanical feasibility and the impact on mmWave performance. Without this approach, there is little chance to meet the window of opportunity to bring these ICs to market.
Moving forward, the challenge will be lowering the cost of test for mmWave devices. Today it is conceivable that the cost of test is 80% the cost of the IC for first-generation 5G devices. This situation cannot be sustained. As volumes of mmWave devices increase, there will be some natural economies of scale that drive down costs. More efficiently designed test cells and interface hardware will need to complement the economies of scale to bring the cost of test down to palatable percentages within the coming years.
Contacted production test at mmWave frequencies is the norm today. Metal transmission lines connect the tester to the DUT for up to 80 GHz testing. Ultimately, these mmWave devices are connected to an antenna module that communicates wirelessly to the outside world. However, space constraints and cost are pushing vendors to integrate antenna arrays directly into the semiconductor package or substrate. The emergence of antenna in package (AiP) is changing the landscape of test. AiP requires an over the air (OTA) mmWave interface to the DUT. Embedding a wireless interface in the ATE environment is nontrivial. Test vendors, such as Cohu, are embedding horn antennae and patch antennae into the contactor, test head and/or the handler interface.
As more devices move into the mmWave frequency bands, test infrastructure may morph into very different form factors. Although a few devices incorporate mmWave components, today the trend toward the connected world will continue to push the limits of existing speeds and bandwidths. This trend will require new technologies in test to confirm functionality and reliability of these devices. Traditional transmission line topologies will be obsolete. Waveguides will replace cabling and PCB traces, antennas will replace contacted interfaces. It is foreseeable that semiconductor devices will no longer incorporate physical contacts to provide power or to communicate with the external environment. What will the ATE test environment look like then?
Version: February 2019
Author(s): Jason Mroczkowski and Dan Campion, Interface Solutions Group
Featured In: Chip Scale Review January-February 2019