The world is still analog

The Wilmington facility includes a complete wafer-fab line that produces engineering prototypes and production wafers. The assembly of engineering ICs takes place in Wilmington or at a third-party contractor's facility. Production wafers are assembled at other facilities around the world.

Before an op amp, digital-to-analog converter (DAC), analog-to-digital converter (ADC), audio coder/decoder (CODEC), or RF IC is ready for production, it must endure weeks of tests by product engineers. These engineers test devices for every specification that's intended for a data sheet, and they also perform other tests they deem necessary. Test results often become the "typical" performance characteristics published in data sheets. Upon completing an evaluation, product engineers write a 30-to-40-page test report. (My first job after college was as a product engineer at Analog Devices. Read a summary of my experience in "I was an Analog Devices engineer.")

Each preproduction IC needs an evaluation board that provides engineers with access to signal pins and programming registers. Boards often contain communication ports such as USB to communicate with a PC. A product engineer is responsible for developing an evaluation board. During my visit to Wilmington, I heard a common refrain about performing a product evaluation: "Is it the board or is it the device under test that's causing the problem?" Often, the engineers must go through a process of elimination to locate a problem's source.

When they initially test a product, the engineers measure the DUT with manually controlled bench instruments. Once they're convinced that the part functions properly, they often run automated tests with in-house test software on the bench or connect the evaluation boards to automated test racks. Finally, they use production ATE systems to perform tests on devices over a range of temperatures and power-supply voltages. The ATE systems are built by a former division of Analog Devices. The company now focuses entirely on the design and manufacture of ICs.

Francisco Santos is the product development engineering manager for the high-speed amplifier group. His team of engineers evaluates products such as high-speed, low-distortion amplifiers, video filters, active RF splitters, cable drivers, equalizers, and ADSL line drivers.

Flanked by an amplifier THD test rack, Francisco Santos oversees evaluations of op amps, video filters, active RF splitters, equalizers, and ADSL line drivers.

Engineers in Santos' group evaluated the AD8099, a low-noise, low-distortion, high-speed op amp that has a gain-bandwidth product of 3.8 GHz. Because Analog Devices develops numerous amplifiers each year, engineers have built several automated test systems for the engineering lab. One of these systems measures total-harmonic distortion (THD), an important specification for op amps. "Distortion levels for our low-distortion op amps can run 140 dB below the fundamental frequency of a test signal," said Santos, "so the noise floor of our instruments is very important. We don't want to be measuring the noise in the instrument."

When performed manually, harmonic-distortion measurements can take weeks, but engineers such as Greg DiSanto now run the test in just hours with an automated test station in the lab (Figure 1). DiSanto characterizes THD versus frequency, amplitude, supply voltage, and common-mode voltage. This station uses a Stanford Research signal source to produce test frequencies up to 200 kHz and a Rohde & Schwarz unit at higher frequencies. When used to characterize video amplifiers, both signal sources produce 2-V_p-p sine waves set to the 3-dB point of the selected low-pass filter. Low-pass filters from Allen Avionics—connected through 18-GHz RF switches from Keithley Instruments—remove harmonics before they reach the DUT. More RF switches connect the DUT's output to filters, which enables an Agilent Technologies spectrum analyzer to measure distortion caused by second and third harmonics.

Figure 1.  An automated test system measures THD in high-speed op amps. View a larger image of this figure.

Figure 2.  An automated distortion test system for video splitters uses 135 input signals.

To characterize CSO and CTB, Ciarlone measures intermodulation products generated by the 135 channels. The band-stop filter on the DUT's output removes out-of-band carriers. He measures XMOD by modulating all channels except the one of interest and measuring the resulting "spill over" in the unmodulated channel band. "We can get the data we need in a fraction of the time with all channels on," noted Ciarlone, "and fully evaluate a splitter."

Engineers in the high-speed amplifier group use Agilent Vee to control both the THD and splitter testers. They have automated several other measurements such as noise-spectral density using spectrum analyzers from Agilent and Rohde & Schwarz, and video-filter frequency response and group delay with an Agilent network analyzer and Keithley RF switch.

Product engineer Gina Colangelo evaluates high-speed DACs prior to production.

Munson's most recently released product is the AD9736, a 1.2-Gsample/s, 14-bit DAC. He's currently evaluating a higher-speed part. He measures noise spectral density (NSD), THD, spurious-free dynamic range (SFDR), intermodulation distortion (IMD), adjacent-channel power, bit-to-bit skews, linearity, and power consumption. Test signals include a single-tone and dual-tone sine wave. An Agilent ParBERT generates the signals in digital form, and it provides 14 pairs of differential signals.

When he evaluated the AD9736, Munson tested the devices at 600 Msamples/s, 800 Msamples/s, 1 Gsample/s, and 1.2 Gsamples/s. At 1 Gsample/s, Munson programs the ParBERT to sweep from DC to 490 MHz, just under the Nyquist frequency of 500 MHz.

"When you evaluate DACs that run at speeds over 1 Gsample/s," said Munson, "the layout of your evaluation board is critical. It's often difficult to isolate problems that come from the board or from the device under test. An evaluation board requires good power-supply decoupling, and its digital traces need to be separated from analog traces."

DACs such as the AD9736 have differential analog outputs, which minimize system errors caused by ground loops. That's fine for engineers who design the devices into systems, but it makes testing more difficult than with single-ended outputs. Munson uses transformer loads instead of amplifier loads in his evaluations.

"Transformer loads make it easier to evaluate a DAC because they convert differential outputs to single-ended outputs without having to worry about the additional nonlinearities of an amplifier output stage," he said. Munson found that he needs more than one transformer to fully test a DAC because of the operating bandwidth limitations of each transformer.

For her part, Colangelo evaluates DACs such as the AD9779, which use single-ended CMOS inputs. She performs the same measurements as Munson. "Some customers are not yet comfortable with the high-speed LVDS interface DACs so they choose a CMOS interface DAC with added digital functionality," she said. The input data rate can run at speeds up to 300 Msamples/s. To generate the single-ended test signals, Colangelo uses a pattern generator developed at Analog Devices.

Used primarily in cellular base stations, the AD9779 contains two DACs to generate I/Q modulated signals. Users can power down parts of the device to conserve power. For example, they can power down the DAC cores when a handset isn't transmitting.

The AD9779 uses digital filters to smooth the device's analog output. The digital filters require their own clocks, but those clocks can add distortion to a DAC's analog output. Isolating clocks from analog circuits is a challenge for both IC designers and product engineers. "We have to test not only the device," said Colangelo, "but the evaluation board, too. If you power down the DAC section of a device but leave the digital control section powered and measure its output with a spectrum analyzer, you shouldn't see any spurious signals. If you do, you may have a board issue."

When they initially evaluate a DAC, Munson and Colangelo perform bench tests manually. "We need to first understand how the device works," said Colangelo. "After that, we can automate our testing. We must also make sure that the spectrum analyzer isn't adding distortion to our measurements." To get the best possible performance, they use two spectrum analyzers to evaluate their DACs. "We use a Rohde & Schwarz spectrum analyzer up to 100 MHz," noted Colangelo. "Above 100 MHz, we switch to an Agilent instrument."

After Munson and Colangelo are confident that the part works and that their evaluations don't add distortion, they run a series of automated measurements. They control the pattern generator and spectrum analyzers using software written in National Instruments' LabView. Following bench tests, Munson and Colangelo will use an ATE system to make many of the same measurements over a range of temperature and power-supply voltages.

ADCs are another core product for Analog Devices, and product engineer Chris Carney evaluates them. He tests ADCs that use both differential LVDS and single-ended CMOS digital outputs.

"Typically, 350 Msamples/s is the switchover point from CMOS to LVDS," said Carney, "but some customers want LVDS outputs even at lower speeds." Customers who use ADCs at speeds of 100 Msamples/s to 200 Msamples/s may prefer LVDS because the differential outputs use a smaller voltage swing.

Carney's ADC evaluation boards connect to a FIFO memory board. Two versions, with 16 kbytes and 32 kbytes of memory, let him run an ADC at full speed and analyze the data offline after transferring the data to a PC. Like Munson and Colangelo, Carney begins his evaluations running manual tests on the bench before automating his measurements.

His FIFO board works with in-house ADC test software called LabAlyzer, which is an executable written in LabView. With LabAlyzer, Carney can configure an ADC, capture data, and perform FFTs to measure distortion and integral nonlinearity. One of his tasks is to control a register that adjusts the ADC's input bias voltage. Once he finds the optimal bias voltage, design engineers can set that voltage in silicon for production devices.

Product engineer Chirag Patel evaluates audio CODECs for use in automotive sound systems.

Starting with an evaluation board, Patel configures the CODEC for a specific mode of operation. He configures the sample rate, serial data format, and volume by writing to registers through a serial peripheral interface (SPI) port. The device has 18 user registers and several others for in-house diagnostics only. The evaluation board communicates with a PC through a USB port.

When first debugging a new part, Patel runs into some of the same issues as his colleagues Munson and Colangelo—identifying the source of noise. "It's a process of elimination," noted Patel. "If I see noise on the power-supply lines, I use an external power supply instead of the evaluation board's supply."

Patel measures IMD, THD+noise, linearity, signal-to-noise ratio (SNR), and crosstalk by exciting the CODECs with single and multitone signals from an Audio Precision tester. In a THD+noise test, he typically uses a 1-kHz sine wave with an amplitude 1 dB below the device's maximum input level, and he measures second and third-order harmonics with the audio tester.

Patel initially receives about 50 devices for evaluation and debug. As part of his bench evaluation, Patel checks for device functionality in all possible modes of operation. After the bench tests, Patel uses an ATE system to further characterize the device, looking for critical digital-interface timing limits.

Digital timing characterization may involve skewing digital signals with respect to each other, which is how Patel measures setup-and-hold time. He characterizes CODECs over the specified temperature range, power-supply voltages, and variations in the wafer-fabrication process. If the parts meet specifications, he evaluates about 500 parts on an ATE system for statistical evaluations. From the statistics, he can assign typical and guaranteed values for the part's published data sheet. The statistical measurements include the analog characteristics THD+noise and SNR.

"We typically hold back about 50% of a preproduction run of partially processed wafers in case we need to make changes," said Patel. "If changes are simply to the logic circuits of the CODEC, then new prototype parts can be ready in about three weeks. If changes are required to the device's analog circuits, a change may take as long as 12 weeks."

Analog Devices also manufactures RF and optical components developed and tested by engineers in the RF and wireless (RFW) group. Senior product engineer Tom Kelly evaluates RF products such as power detectors, amplifiers, multipliers, and modulators as well as optical components such as log detectors. The RFW group has several automated test benches, one of which Kelly used to test the AD8349, a 700-MHz to 2.7-GHz quadrature modulator used in GSM and CDMA cell phones.

Figure 3.  The spectrum analyzer measures noise, adjacent channel leakage ratio (ACLR), and sidebands in an I/Q modulator.

Figure 4.  An adjacent channel leakage measurement shows less adjacent and alternant-channel noise in the newer part (blue trace).

Figure 5.  The output spectrum of an I/O modulator shows undesired frequency components.

Kelly is concerned about how calibration affects a modulator's I and Q baseband inputs. If the signals aren't equal in amplitude and in quadrature, Kelly will see an undesired sideband. Just 1° of phase error, even with perfect amplitude match, causes a –40-dBc undesired sideband. A 1° phase and 0.5-dB amplitude error generates a –30-dBc undesired sideband.

Engineers at Analog Devices spend weeks evaluating new IC designs on the bench, on automated test stations in the lab, and on production ATE systems. A product engineer must approve a product for production, and he or she provides valuable feedback to designers.

The May 2005 issue of Test & Measurement Worldincluded an interview Martin Rowe conducted with a product engineer at Analog Devices' design center in Beaverton, OR. See "Northwest passage," www.reed-electronics.com/tmworld/article/CA528092.