Most industry observers agree that mixed-signal devices are the wave of the future. With growing product convergence of sound, video, and 3-D graphics and computing, it’s not surprising that mixed-signal testing is the fastest growing segment of the IC industry. As a result, test professionals will likely find themselves facing mixed-signal devices more and more often.
The challenge of mixed-signal devices is to accurately test the digital and analog functions completely along with the complex interactions between the two. These interactions may be layout-design-related such as cross talk and noise or more logic-design-related as analog and digital functions become more embedded.
The earliest mixed-signal devices were predominately analog or digital. In cases where 98% of a design may have been digital, test efforts were focused on the digital functions with only a single checkpoint for the 2% of the device that was analog.
Today, the digital-to-analog ratio on mixed-signal devices is more balanced, making it imperative to focus equally on both digital and analog circuitry. Highly embedded designs make some of the circuitry inaccessible for testing without the proper equipment and the know-how to precondition the device properly. This requires collaboration between the device designer and test engineer.
Evaluation vs Comparison
The essence of mixed-signal testing is the evaluation of data using mathematical formulas in contrast to the pass/fail approach used in digital logic-only testing. Ultimately, digital logic testing entails sourcing digital input data and then comparing the output data to a model of the device function or pattern.
In mixed-signal testing, the raw device output data cannot be compared directly to a model to determine a pass/fail condition. It first must be manipulated using mathematical and statistical functions to arrive at a final result, which then can be compared to a specification. It is these characteristics that determine a pass or fail rather than its functionality.
Getting Equipped
Given the digital/analog balance and integration of current mixed-signal technology, it’s becoming less feasible from a cost and cycle-time standpoint to use an analog tester for the analog-only portion and a digital tester for digital-only functions. The test engineer has to learn two completely different tester languages, develop two incompatible sets of hardware, and write two different test programs. In manufacturing, the company would have to buy and support two tester platforms, taking up valuable floor space as well as testing each device twice. Even with that, a complete test could not be performed.
The interaction functions between the two blocks not addressed in the two-system approach are more pervasive now and cannot be overlooked in testing. A high-performance mixed-signal test system is required for two reasons:
Any cross talk, noise, or other second-order effect can only be replicated on a mixed-signal tester.
With today’s high level of embedded integration, it’s likely that digital patterns would be used to precondition the analog tests or vice versa.
In a mixed-signal test system, the source and capture hardware sections of the tester provide information in either analog or digital form. The digital section drives and compares digital signals. The analog analysis hardware section evaluates the captured data as a signal with a digital signal processor (DSP) used to extract signal characteristics in both the time domain and the frequency domain.
Starting at the Source
The analog side of the signal source in a mixed-signal tester is an arbitrary waveform generator that works like a memory behind a digital-to-analog converter (DAC) within the system. A numeric model of the waveform, the sample set, is stored in the source memory and repeatedly cycled out to the DAC. This sample set generates a sequence of voltage levels that reproduces the wave shape in analog form.
Getting the source to provide proper input to the device requires creating a periodic sample set. Because the source repeatedly loops on the sample set to generate a continuous waveform, the sample set must be configured so that the loopback generates a smooth transition. The faster that data can be clocked through the tester DAC, the more sample points that can be generated within a given period for a smoother sine wave. Amplitude resolution also is important in generating enough discrete levels to ensure that the output waveform will result in maximum test accuracy.
The appropriate number of samples and sample frequency is determined with respect to the test tone to facilitate proper signal analysis. Fast Fourier transform (FFT) produces the spectrum that is used to calculate the dynamic specifications, requiring 2n sample points and an integer number of cycles:
Sample rate/test tone frequency = # points/# cycles
For example, in testing an 8-bit DAC device with a 28-MHz sample rate, a 0 to 5-V output range, and a test tone of 95.7 kHz, a pattern that generates 2,048 points must contain seven cycles. The sampling rate of the tester usually sets the same rate (28 MHz) and captures the same number of points (2,048).
On a mixed-signal tester, some digital source data typically is stored in dedicated source memory where it represents a test tone rather than 1s and 0s and a digital test vector pattern. The stored data is the input to the arbitrary waveform generator that drives the analog inputs. Data doesn’t come from the pattern memory and, as a result, does not represent a logic function. On the other hand, functional digital data driven into a digital input on the DUT does come from pattern memory.
In our 8-bit DAC example, you would drive a sample set of 1s and 0s on the input. The greater the number of sample points in a given time, the smoother the shape of the sine wave that sets the stage for the capture phase of the process.
Capture Is Key
The signal capture hardware of a mixed-signal tester is a waveform digitizer with analog and digital inputs. In the analog portion, an analog-to-digital converter (ADC) samples and converts the analog signal from the device into a digital code that corresponds to a numeric value.
The tester digitizer should have a resolution of at least two more bits than the device being measured (in the case of our DAC example, 10 bits) to ensure the noise floor is low enough to avoid masking out any frequencies of interest. In the example, each of the 2,048 sample points, as calculated by FFT equations, is digitized and stored in the capture memory.
Sampling the output signal is a function of required resolution: the greater the number of samples, the more precise the digital representation. As with the source, it’s beneficial to get the points in the sample set close together. The faster that data can be clocked through the tester ADC, the greater the precision of the acquired data set. Techniques such as under-sampling (used when the sampling rate is less than two times the frequency of the signal of interest when a slow tester is used to test a fast device) or over-sampling (when the number of samples per cycle is small) can be used to increase effective resolution.
As with the source, the digital side of capture is similar in function to the compare data on a logic tester. Digital values are stored in capture memory and analyzed as a signal. Digital capture of a digital pattern output is compared to expected data.
Signal Analysis: The Final Hurdle
In the signal-analysis portion of mixed-signal test, the tester’s DSP processes data from the capture memory (Figure 1). Rather than comparing each bit in each vector against expected data, the DSP processes the entire sample set as a signal. The DSP system doesn’t control measurement hardware—it replaces it.
Using the DSP is like writing a separate subroutine within the test. Results from the subroutine are passed back to the main program as parametric values which then are compared with pass/fail limits in the main program.
To thoroughly test a mixed-signal device, the signal must be analyzed in both the time and frequency domains. Figure 2 illustrates a signal in the time domain where variations in amplitude can be measured over time. In addition to peak amplitude, frequency, and jitter, typical tests in this domain are:
Root mean square (rms)—determines the area under the curve; design dependent.
Integral nonlinearity (INL)—determines the best fit straight line through the voltage end points. The typical limit is ±1 least significant bit (LSB).
Differential nonlinearity (DNL)—measures the difference between actual and theoretical voltage step sizes. An input voltage ramp usually is sufficient to do INL and DNL tests.
Settling time—analyzes the time it takes for output to settle.
Gain—measures the difference between ideal and actual analog output at full-scale digital input; design dependent.
Figure 3
shows the same signal in the frequency domain, which represents variations in power over frequency. Testing in this domain often involves evaluating characteristics relating to signal shape or purity since some features of signal shape are more easily quantified in the frequency domain than in its time-domain counterpart. There are three typical tests:
The Spurious Free Dynamic Range (SFDR) test compares the power of the fundamental frequency (95.7 kHz) to the highest power of all the harmonic frequencies (191.4 kHz, 287.1 kHz…). A typical result would be 54 dB for a DAC device such as our example.
The Total Harmonic Distortion (THD) test compares the power of the fundamental frequency to the total power of the harmonic frequencies up to a certain order, usually to the fifth harmonic. A typical result would be 52 dB.
The Signal-to-Noise Ratio (SNR) test compares the power of the fundamental frequency to the total power of noise and is taken at a certain bandwidth, such as 500 kHz. The noise energy contains all frequencies up to the bandwidth minus the energies of 0 Hz, the fundamental frequency, and all harmonic frequencies. A typical result would be 50 dB.
Although each of these tests has been described separately, design and test engineers must determine if the DUT can be tested in that manner. Design implementation, end application, time schedules, tester capability, and cost are key considerations.
Hands-On Testing
The process of sourcing, capturing, and analyzing data in mixed-signal testing is markedly different than in the worlds of digital logic or pure analog. In addition to the two blocks of a mixed-signal device, there is the interaction between them that must be tested as well.
Unlike digital testing, there are no industry-accepted programming tools available to automate mixed-signal testing. That leaves it to the test engineer, who must take a hands-on approach to this type of testing.
While it’s necessary to learn the nuances of this approach to succeed in mixed-signal test, having the right expertise is just part of the challenge. You also must have the right equipment. Moving forward, as succeeding generations of mixed-signal devices become ever more complex, the combination of expertise and equipment will be even more crucial to achieve quality test results.
About the Authors
David Ganapol is vice president of engineering at Artest. Mr. Ganapol has been in the semiconductor test and ATE industries for more than 25 years. Previously, he was the test engineering manager for VLSI Technology’s Worldwide Assembly and Test Operations. (408) 731-8778, ext.112.
Keith Imai, an industry veteran with 20 years of experience, is the director of sales and marketing at Artest. Mr. Imai also has served as marketing manager and engineering section head for the Advanced Computing Division at VLSI Technology. (408) 731-8767.
Copyright 1999 Nelson Publishing Inc.
November 1999
|