At the core of many precision ATE systems is an A/D converter, whose accuracy and performance defines the system's overall accuracy and performance.
Precision ATE systems require high resolution A/D converters to digitise analogue signals and the analogue signal chain requires excellent DC specifications, such as offset, gain and linearity.
In order to achieve ppm level resolution and accuracy, precision ATE systems are calibrated digitally to null out system level offset and gain errors. As a result, system accuracy is often limited by errors that cannot be suppressed by infrequent calibration and system designers may be more concerned with potential drift of key parameters than they are with their static values. For example, alongside ppm level accuracy at a fixed temperature, systems may also need sub ppm/°C drift accuracy over a wide temperature range.
A/D converter linearity is of critical concern for overall system accuracy and determined by complex interactions between the analogue input signal and the converter's design and architecture. Non linearity errors are difficult to calibrate at the system level, since they vary from one digital code to another and because they may be a strong function of temperature.
To meet these challenges, a family of 20bit SAR converters from Linear provides performance and accuracy, simplifying the design of high precision ATE systems.
SAR A/D converters are characterised by their ability to acquire a precise snapshot in time of an analogue input signal and to complete an A/D conversion within one clock cycle. They excel at asynchronous 'start and go' operations and are easy to use because the conversion result is available within the same clock cycle. The ability to produce accurate conversion results with no cycle latency, even after long idle periods, makes such parts ideal for precision ATE systems.
The SAR A/D algorithm is based on a binary search principle. The analogue input is sampled onto a capacitor and compared sequentially to fractions of a reference voltage selected by the SAR algorithm. The converter comprises three components: a capacitor based D/A converter (CDAC); a fast low noise comparator circuit; and a successive approximation register. The INL performance of a conventional SAR converter may be limited by finite matching accuracy of individual capacitors in the CDAC and precision SAR converters often employ analogue or digital trimming techniques to improve this. However, as temperature varies and package and board stress is applied, CDAC capacitor matching degrades and may limit the converter's linearity.
Breakthrough performance
The LTC2378-20 is the flagship product in a family of pin and software compatible devices featuring up to 20bit resolution with no missing codes and an SNR of up to 104dB at sample rates of up to 2Msample/s. DC precision is particularly impressive: the converter's integral nonlinearity (INL) errors are typically less than 0.5ppm and guaranteed to be less than 2ppm for all codes over temperatures ranging from -40 to 85°C. The maximum offset error is 13ppm with 0.007ppm/°C drift, while the gain error is 10ppm with 0.05ppm/°C drift. Power consumption ranges from 5.3mW at 250ksample/s to 21mW at 1Msample/s.
The guaranteed linearity and accuracy of the LTC2378-20 is a 'game changer' for many precision ATE systems. The device has been designed using a proprietary architecture that ensures linearity and minimises its sensitivity to changes in temperature and other operating conditions. As a result, an unprecedented 2ppm INL specification is guaranteed over the entire operating temperature range.
The LTC2378-20 achieves its performance by implementing a proprietary architecture that makes INL independent of CDAC capacitor mismatch. This makes it robust to the temperature variations and package stress found in industrial environments. Power consumption is proportional to the sampling rate, so they consume only microwatts when operated at 1ksample/s.
Accuracy and speed
High channel count ATE systems may use slow A/D converter architectures for precision DC measurements, with multiplexers allowing a single meter to service many inputs. While conversion time can be traded against resolution, measurement resolution is often less than 16bit at sample rates of more than 100ksample/s.
The LTC2378-20 can take 1million readings per second, with 2.3ppm noise resolution. Results from multiple readings of the same analogue signal may be combined digitally to improve the noise resolution, yielding performance exceeding that of multislope converters. For example, by averaging blocks of 10 samples, the LTC2378-20 effectively operates at 1Msample/s/10=100ksample/s with a 0.7ppm noise resolution (114dB SNR).
Delta-sigma and multislope converters may be configured to average an input signal during an observation/integration period to suppress noise and interference. An observation period of 100ms is often used to suppress 50Hz and 60Hz line frequency interference, resulting in a throughput of 10sample/s. Accordingly, it takes a full second to service 10 multiplexed channels with one multislope converter. Fig 1 shows an LTC2378-20 operating at 102.4ksample/s, configured with a multiplexer circuit to simultaneously measure all 10 signals (interleaved) during the 100ms observation period. While preserving the suppression of line frequency interference, throughput is increased by the multiplexing factor (here 10, but can be higher), bringing higher productivity.
In this example, noise resolution is increased by averaging across 1024 samples taken from each channel during the observation period, providing 22bit of noise resolution (0.07ppm or 70nV rms). The averaging operation can be performed with a simple adder, allowing the LTC2378-20 to increase measurement speed significantly, while maintaining the advantages of prior architectures.
Because one LTC2378-20 can potentially replace several discrete components required for a multislope design, a degree of design freedom cost, board space and channel count to be balanced. Replacing a multiplexed meter with one or more LTC2378-20s can shrink system size, reduce power and solution cost, and increase speed. Because the device can operate in its native mode as a Nyquist converter at up to 1Msample/s, one LTC2378-20 is ideal for use in systems that would otherwise require more than one type of A/D converter.
Simplifying the signal chain
Some ATE systems may require signals to be evaluated with great precision and with some bandwidth. More bandwidth implies more noise, so such systems typically require two digital data streams: one with low noise, low bandwidth and high accuracy; and one with higher noise, higher bandwidth and lower accuracy.
The conventional approach is to use separate converters for each stream, with each converter optimised for accuracy or bandwidth/noise (fig 2). The LTC2378-20 is optimised for both objectives and may be used in both data streams.
When oversampling a SAR A/D converter, antialiasing requirements are relaxed, as they are for delta-sigma devices. However, SAR converters have not matched the linearity of delta-sigmas. That has changed and the INL of the LTC2378-20 is better than that of delta-sigma devices. This opens many new exciting opportunities in ATE applications, see fig 3.
By replacing two very different and separately optimised A/D converters with one SAR converter, system design can be simplified.
Atsushi Kawamoto, design manager, Jesper Steensgaard, staff scientist, and Heemin Yang, design section leader, are with Linear Technology.
