This is the second article of a three-part series discussing the likely impacts of next-generation consumer wireless semiconductor devices on instrumentation performance requirements. The first article, which appeared in the October issue of EE-Evaluation Engineering, examined the structural differences between device generations to suggest the likely impact presuming use of existing RF ATE.
Issues presented in the first article focused on the forces driving the wireless industry. Increases in consumer wireless are fueling unprecedented unit volume growth across a wide variety of cellular and connectivity standards. Some sectors are maturing into a high-volume, low-price phase. This motivates innovation, which in the semiconductor business usually means integration.
Innovation also is occurring in new and evolved wireless standards. This generates new RF parametric technical requirements. When both innovations collide, new requirements evolve for high-speed digital interface circuits. Device characterization and release processes are complex due to continued die-shrinks and the demand for high quality.
The existing installed base of RF ATE second-generation (2G) testers is designed primarily for low pin-count transceivers having minimal digital and analog test requirements. This article discusses new requirements driving the need for improved technical capabilities.
RF semiconductor devices are changing to match the needs of increasingly innovative wireless standards. In turn, device functionality and integration drive core parametric, configuration, and usage trends for RF ATE.
Signal bandwidths and carrier frequencies supporting greater data throughput rates are increasing steadily. Changing packaging requirements and the need for yield improvement are driving test-margin reductions.
Integration is adding high-speed digital circuits to traditional transceiver RFICs, which previously had digital functions no more complicated than register controls interfaced via three-wire serial. Given the increased complexity of the devices, device characterization requires more tests over wider operating ranges. Increased unit counts and multiple production lots also are needed to facilitate statistical correlation.
Frequency and Bandwidth
There is a fundamental trade-off between high data throughput expressed in megabits per second and the bandwidth available for new transmitters within the scarce RF spectrum. Governments have authorized specific uses of radio spectrum for almost a century, so the most highly desired frequencies, those low frequencies that propagate farthest, are already assigned.
Those frequencies also tend to have narrow bandwidth definitions, so many existing allocations must be moved to accommodate new wider bandwidth allocations. This is driving the carrier frequencies higher in new standards relative to prior standards.
Wireless carrier frequencies and bandwidths have increased steadily along a logarithmic trajectory for two decades. Figure 1 shows the first year of a major metropolitan introduction rather than the year of invention, which was 1967 for AMPS and analog cellular for the United States and Canada. Today, wireless usage is segmenting into a variety of distance and data throughput needs.
Figure 1. Progression of Wireless Carrier Frequencies and Production Bandwidths Since 1980
The inherent convenience of wireless mobility is driving a variety of new market uses including ultra-low-power, narrowband standards such as Zigbee. This proliferation of standards and signal bandwidths will continue through the next decade.
Test capability must keep pace with new standards. Next-generation RF ATE must follow both the carrier and bandwidth trends. In comparison, 2G testers typically have maximum carrier frequencies of 6 GHz and maximum receiver bandwidth of 20 MHz, which are inadequate for today's emerging wireless device needs.
Wireless segments are emerging along distance and data throughput metrics (Figure 2). Standards supporting the highest data rates today tend to have shorter transmission distances. For example, the spectrum-sharing standard ultra-wideband (UWB) has a regulated maximum of -42 dBm/MHz, more than four orders of magnitude less than a milliwatt. This compares to a few watts for cellular transmissions of about 1 MHz.
Figure 2. Progression of Wireless Segments in Terms of Distance and Data Rate
UWB operates best around a 10-meter perimeter but achieves well beyond 100 Mb/s throughput. For the coming decade, the interesting battleground is the high distance to 10 km and high data rate area of >50 Mb/s where WiMAX and fourth-generation (4G) cellular will compete for both stationary and mobile usage.
Increased connectivity functions within handsets are leading to cross-standard integration. Separate Bluetooth and wireless LAN RFSOCs have been available for a few years. Now those devices are being integrated together.
In addition, FM transmitters will integrate to facilitate automotive use of handset-resident music files. Handset connectivity integration trends will likely continue past audio and GPS functions to video, such as transmission of DVD outputs and handset digital broadcast to larger media like HD televisions, computers, and in-vehicle entertainment systems.
Next-generation RF test must support a wide range of bandwidths and frequencies from 76 to 108 MHz for FM to the 5.9-GHz carrier of WLAN and 11-GHz carrier of UWB. Digital pin-counts are increasing noticeably with some chips having more digital pins on a single device than the 64 to 128 digital pins on a typical 2G tester configuration. Supporting cost reduction through multisite for these integrated devices is not feasible on RF-focused 2G testers.
The 11-GHz carrier of UWB is just one of several requirements causing problems for existing RF ATE. The 512-MHz bandwidth creates misalignments in RF instrumentation as well as analog and digital requirements.
First, RF receivers in 2G testers typically have about 20 MHz of bandwidth, and newer RF testers support 40 MHz. While this easily accommodates spectral mask measurement, assuming spectrum analyzer-like multiple pass measurements across frequency, it cannot demodulate the RF signal. As a result, calculation of error vector magnitude (EVM) and phase error on a 2G tester will require special handling, sometimes with GPIB instruments.
Second, a tester's analog requirements must serve a bandwidth of at least half the RF bandwidth. 2G testers were built during the late 1990s for AC requirements in the single-digit megahertz range. Supporting 256 MHz and higher for characterization tests is not feasible without increased costs and test time. Supporting both modulation and demodulation for transceivers requires analog generators and digitizers of adequate frequency range.
For some UWB designs, it is possible to provide adequate device fault coverage with a CW-only test process. However, ensuring correlation to CW tests requires design verification and process correlation to both modulated performance metrics and strict regulatory requirements.
Cost of test and time-to-market issues also must be considered in what is likely to be a very competitive, hyper-growth market segment. Some analysts have predicted that the UWB growth inflection will be faster than any prior wireless standard.
Further, the average selling prices will fall rapidly as both start-ups and major manufacturers wrestle for, or against, comparative advantages. Accordingly, the capability to rapidly deploy low-cost, efficient fault coverage against correlated modulation tests is essential in this potentially rewarding new application of RF technology.
Designers of consumer devices are demanding more stringent controls on device performance and conformance to global rather than merely regional governmental requirements. In turn, this is driving an increased focus on modulation metrics.
Measuring the output spectrum of GSM/EDGE, WCDMA, and WLAN devices is common today, and the same is expected of WiMAX. New standards such as WiMAX and improved standards like the new 802.11n WLAN are adding additional complexity to the devices.
The disappearance of historical test access points necessitates greater emphasis on system-level performance metrics such as EVM and spectral occupancy, for which measurements 2G testers lack needed sensitivity and second- and third-order intermodulation performance.
Improved process technologies have steadily improved transit frequency and maximum frequency of oscillation for RF semiconductors. Designers of new standards such as 802.11n are taking advantage of the enhanced capability to improve characteristics such as device phase noise. Both the frequency and the phase noise requirements of reference clocks have increased.
Using a digital pin for the device reference input no longer is adequate for devices supporting -145 dBc/Hz at a 10-kHz offset. Nor is direct local oscillator substitution feasible because the test access is not available in RFSOC packaging.
Device interface board (DIB) circuits have been used in the past at the cost of board space, which is at a premium starting at single and dual site for 2G testers. Commercial frequency dividers and fanout amplifiers can routinely reach the -135 dBc/Hz range, but achieving -145-dBc/Hz levels is nontrivial. As a result, adequate reference signal distribution will be left as an exercise to the user without improved ATE instrumentation or the addition of costly GPIB instrumentation.
Power, or work per unit time, is a simple concept, but it can be very difficult to measure accurately. At the same time, level accuracy is critical to the operation of cellular handset networks.
First, transmit power is regulated. Second, CDMA cellular networks rely on the capability to control handset power to maximize the number of users supportable within a single base-station sector.
Handset transmit levels are dynamically controlled to be a specific power level as seen by the base-station receiver, which is necessarily receiving signals from multiple handsets. One handset transmitting 3 dB above the others will drastically reduce the channel capacity of that base-station sector; in some cases, capacity is reduced by half.
Third, handsets use multiple active and passive RF components between the antenna and the digital baseband circuits. Each component in the handset RF circuit has a gain/loss budget. The actual value depends on the nominal value of the component's insertion characteristic and the impedance match between itself and the adjacent components. The passives, amplifiers, and transceiver all have a specification and related test margin requirements.
A laboratory-grade power meter has a calibration process traceable to international standards via working standards utilized for calibration at the place of manufacture. The working standards are calibrated with one or a very limited number of calibration sensors that physically make an annual journey to a recognized standards laboratory such as NIST.
Power sensors are designed to have impedance characteristics very close to 50 Ω, and the physical distance from the coaxial connector to the microwave detector element is less than 10 mm. The design deliberately maximizes the conditions for a convenient, accurate measurement. Even so, worst-case accuracy is routinely above the 10% range with mismatch effects requiring a vector error correction that a power sensor/meter cannot provide directly, accounting for 4% to 6% of the error given a better than average impedance match.
An ATE environment makes accurate level measurement difficult. First, the devices do not have coaxial connections. Second the physical distance to the calibrated, coaxial ports of the RF instruments is in the 15-cm to 30-cm range, a condition that exacerbates the level variation effect vs. frequency due to mismatch signal reflections.
Third, any component that is not impedance matched to an adjacent component will cause a mismatch and a signal reflection. This includes device matching circuits, baluns, adapters, and cables. For example, mismatch variations within a cable are possible due to crimping of the backside interface of the cable's connectors.
Fourth, DIB cables and connectors typically are cheaper versions using Teflon dielectrics that expand and contract with temperature and humidity. This creates a varying impedance environment, and the coaxial mating dimensions are forced closer together or farther apart.
Finally, the balun mentioned earlier converts 50 Ω to the 100-Ω differential needed by almost all recent transceiver designs. Eliminating reflections is not practical. Calibrating through a balun can be difficult, even for experts, because there is an inherent choice to calibrate to the balun match or to the device-socket interface. In sum, almost every aspect of workcell integration hampers accurate RF measurement.
RF probe is becoming common for both RFSIP transceivers and wafer scale packaging of RFSOC devices in the 3-mm/side range and below. The probe cards in use today also tend to have significant reflected power signals due to the mating transitions from RF cables into the probe card through to the planar wafer contacts.
The problem of mismatch behavior increases vs. frequency. The match behavior of the test instrumentation, DIB circuits, and the device tend to become worse as frequency is increased. This creates a significant yield problem for RFSIPs.
Transceiver die must be tested in an electrical environment that is very different from that of the SIP, which will have performance-changing characteristics such as parasitic reactance and embedded passives. The SIP then is tested at the package level.
Correlating the die-level and SIP package-level tests is challenging enough without error variations from the tester, which typically are in the 0.8- to 1.5-dB range for a 2G tester and DIB. Next-generation RF test needs to conform to a 0.2-dB uncertainty range with worst-case uncertainty of ≤ 0.5 dB. Neglecting the correlation need will impact RFSIP yields.
Adding power amplifiers (PA) to the RFSIP solves three major problems for handset designers. First, next-generation transceivers have 10 to 20 RF connections, on average more than twice that of 2G devices such as a quad-band GSM transceiver. Adding PAs removes many 50- Ω traces from the handset's design and production requirements.
Second, integrating the PA allows techniques such as polar modulation, which can improve PA efficiency by an order of magnitude. Third, the use of embedded passives within the SIP can save loading and yield costs. Loading of discrete passives can cost more than the RFIC testing because the 100 to 200 passives in a handset each cost roughly a third of a penny to place on the handset circuit board.
However, integrating the PA causes test problems. Injecting several watts of RF power into a 2G tester's RF instrument will immediately induce damage. Power-handling circuits can be placed on the DIB, but the physical size and calibration requirements reduce site-count and increase DIB calibration and the potential for damage at overseas production locations. 2G testers cannot handle both the power level and the site-count needed for an economic process.
In recent standards, increased data throughput has come at the expense of increasingly complex coding and modulation standards. The device digital speeds are increasing well beyond the 50- to 200-Mb/s rates supported by 2G testers with digital capabilities most commonly used for simple register writes on 2G transceivers.
Today, devices for WLAN access points and PCs are starting to incorporate PCI Express (PCIe). The DDR2 DRAM interface is being added to devices for both the PC and cellular end-markets.
A new standard, 3G DigRF, allows conversion of a transceiver's analog IQ Tx inputs and Rx outputs to serial digital circuits. The Rx data-packet outputs are intermixed with control packets, and both are asynchronous to the operation of the tester's clock. So, not only must the packets be time aligned, they also must be decoded prior to sorting.
Further, the digital data is not repeatable because the device is internally time-sampling an inherently analog process. Its phase is related to the nonlinear product of the input RF signal and the device's internal RF local oscillator. As a result, there is a real-time requirement for the capture of the digital outputs. This is needed within the digital instrument to preserve the capability to determine jitter and level behavior. Inserting a DIB-mounted FPGA between the device's 3G DigRF data outputs and the digital instrument's pin electronics destroys this information.
3G DigRF is a new, 312-Mb/s standard that eliminates the need for analog front-end chips (Figure 3). The traditional analog I and Q signals are converted to serial streams of encoded digital packets so direct connection to the digital baseband chip is possible.
Figure 3. Block Diagram of 3G DigRF Test
Test difficulties in 2G testers follow from the asynchronous, nondeterministic nature of the DigRF standard, which has a 4G version likely to reach 1,248 Mb/s. 2G testers lack the eye-measurement, packet sorting and background decoding, and DSP needed for efficient testing of this interface.
Once captured and decoded, the I and Q data values proceed to a DSP process, which for 2G testers occurs on the host computer. That is, both the movement of the data from instrument memory to the host and the host processing of the data momentarily inhibit further test execution. This increases test time. As device site-count is increased, this part of the test process will achieve near-zero parallel efficiency. Background DSP move and processing clearly are needed for digital as well as analog/RF capture functions.
Characterization on RF ATE
The trend to SOC implementation is occurring concurrently with die shrinks, or perhaps more correctly stated, smaller gate width front-end fabrication processes. Given consumer demands for greater functionality, not all die are shrinking. These dual trends are changing the way that silicon devices are verified and characterized.
Previously, RFICs were routinely verified and certified for production with benchtop equipment. Process variables for RF at 0.35 micron were more static and fewer in number than in processes at 0.13 and below.
Today, integrated RFSOC designs are proceeding at 65 nm and lower. Interfacing multiple cores of integrated digital, analog, RF, and in some cases, DC regulators creates a time-to-market challenge that benchtop testing cannot satisfy. The process variations alone make benchtop characterization impractical.
Thousands of devices are run across multiple process lots to ensure a confidence in the supply chain. With increasingly concentrated consumer device manufacturers demanding and getting commitments for more robust supply chains supporting more than 2 billion RF chips annually, suppliers of RF devices are responding by putting the devices through ATE-resident characterization programs incorporating thousands of tests per device.
Achieving time-to-market now depends on this process. 2G testers lack the performance, such as phase noise and distortion-free dynamic range, needed for validation processes conforming to new standards. For example, when 2G testers were designed, multiple-input multiple-output (MIMO) concepts did not exist. The capabilities of the latest devices in bandwidth and third-order and second-order distortion are well beyond the performance of the 2G tester's internal receivers. Essentially, the 2G tester measures itself instead of the device.
Multisite Drives COT
Next-generation RF devices include more frequency bands and more digital circuits than 2G RFICs. Providing a similar level of fault coverage requires more tests and longer test times. At the same time, the increased pin-counts of the next-generation devices force larger tester configurations. Longer tests on more costly ATE propel a double-double impact on economics. In some cases, the RF or digital pin-count of a single device exceeds the maximum configuration of a 2G tester.
New device types include many more RF and digital pins than prior devices. With growth of MIMO techniques, the trend will likely continue (Figure 4). This growth in pin-count effectively reduces the maximum site-count possible with 2G testers. Consequently, test costs grow due to both test time increases and site-count reduction.
Figure 4. ATE Requirements Needed to Test New Device Types
More sites must be tested simultaneously to reduce these dual drivers of increased test costs. The tester configuration has to be increased well beyond the capability of aging 2G testers.
The parallel efficiency of the tester architecture also needs to be modified to remove the computer host from the process flow. Otherwise, efficiencies will decline to zero for all processes driving the host computer bottleneck.
The physical layout of DIB circuits also is challenging as RF and AC circuits needed to support a single site routinely cover 40 to 60 square centimeters. The physical DIB space of some 2G testers eliminates possibilities beyond dual site.
Elsewhere in the workcell, older handlers need updated mechanics for faster index speeds and plunge accuracy. Some end-users neglect maintenance on handlers, which eventually will suffer due to their mechanical nature. Consequently, the insertion efficiency of poorly maintained handlers can cause misloads and device losses, which can impact costs significantly.
RF devices are changing more rapidly than the RF ATE used to test them. Frequency, bandwidth, modulation, and dynamic-range requirements have gone beyond those envisioned in 2000. With digital speeds faster by an order of magnitude, the processing needs of these test patterns are fundamentally different than in 2G era devices.
The need for improved time to market also is driving characterization requirements. 2G tester parametric performance is well behind that of benchtop equipment, creating a conundrum for project engineers. Verify the device to its capabilities or risk quality issues during the volume market phase.
Trends such as market segmentation, device integration, and performance improvements are a natural result of consumer demands for more capability, convenience, and mobility. Next-generation RF ATE must simultaneously support the diverse performance and economic needs of the new device standards.
The Series Finale
In the December issue, the third article in the series will synthesize the combined needs of 3G ATE and propose a cost structure and system-level requirements based on established multisite economic principles.
About the Author
Ken Harvey is responsible for the RF product line architecture at Teradyne. The 20-year veteran of the ATE industry has a background in RF/microwave and graduated from the University of Akron with a B.S.E.E. and Santa Clara University with an M.B.A. Teradyne, MS 600-2, 600 Riverpark Dr., North Reading, MA 01864, 978-370-3670, e-mail: [email protected]