Optimising Read-Rate in High-Speed ATE

Consider Overall System Test Time

The time that an instrument needs to perform a specific measurement is typically the most important factor in the overall test system timing equation. Another major factor is the time the switching equipment takes to connect to the device-under-test (DUT). You should also consider settling time — the time that you must allow between making configuration changes and the start of a new measurement sequence (both the instrument and the switch system contribute to settling time). These individual time parameters generally add arithmetically, but their relative effect on total system speed varies between applications.

Typical test applications require measuring several points of the DUT using the same instrument, such as a digital multimeter (DMM). Typically, a multiplexer switches test signals from point to point under program control, through relays or solid-state switches. The switching devices normally reside in a mainframe that provides the card cage facility, control lines for the switches, and hardware that implements the various test instructions. The switch system may communicate through several ports, such as external hardware trigger lines, programmable input/output signals, and, most often, remote computer-control connections such as GPIB and RS-232C (Figure 1).

Figure 1. A typical DMM/switch arrangement.

Signal Switching Time

Three factors affect the time it takes the switching element to disconnect one signal and connect another signal to the measurement instrument. These factors add up and determine overall signal switching speed (Figure 2):

Figure 2. Total switching time comprises three components: trigger execution time, actuation time, and settling time.

• Trigger execution time (trigger latency) — the maximum time from activating the trigger source to the start of a switch opening or closing (typically, several hundred ms).

First, Set Up Your Instrument

The time that any instrument takes to make measurements is a major factor in determining overall system throughput. When you’re using a DMM, several key parameters affect measurement speed (many other classes of instrument have essentially similar requirements):

While most DMMs can control many of these parameters, some of the more basic instruments offer no control at all. But the basic parameters remain the same for all DMMs. So, check the level of control that you can exercise, because controlling these parameters is critical for precise measurement timing.

The measurement function (DCV, ACV, and so on) affects measurement rates because each function uses a different measurement technique. DCV measurements are typically the fastest, resistance measurements take longer, and temperature measurements are usually the slowest DC measurement. AC measurements depend on the bandwidth you’re examining (low-frequency measurements are relatively slower). The AC/DC conversion technique that the DMM employs also helps determine AC reading speed — average-responding and electronic true-rms converters are fast, while a thermal converter is hopelessly slow for systems use.

Signal integration time is the time that it takes the A/D converter to sample the input signal. Signal integration time affects the DMM’s usable measurement resolution, the amount of reading noise, and the measurement’s reading rate. The optimum setting for your application depends on your measurement requirements. There is always a trade-off between measurement speed and common-mode or normal-mode noise rejection.

DMM vendors invariably state measurement rates for a fixed measurement range. While some DMMs have fast auto-ranging features, the speed you’ll get is not predictable.

Almost all DMMs provide both hardware and software trigger mechanisms. Hardware triggers result from dedicated trigger lines, general-purpose input/ output lines, and even front panel control commands. Software triggers result from computer-control commands from interfaces such as GPIB or RS-232C.

DMM vendors typically state measurement rates with all filters turned off. Although filtering improves noisy measurements, it also slows measurement rate. Analogue filtering increases the input signal’s settling time by introducing a fixed-time constant that’s typically a multiple of line frequency for DC measurements, or a period that suits frequency response considerations for AC measurements. Digital filtering takes a number of A/D conversions and averages their values before displaying the result. Many DMMs offer two digital filtering modes — a repeating filter and a moving filter (Figure 3). The repeating filter (block-mode) averages a user-defined number of samples, computes one reading, empties its buffer memory of all previous readings, and repeats the process. The moving filter (rolling average) takes a user-defined number of samples, averages them, computes a reading, discards the oldest sample from the buffer, takes a new sample, averages the cumulative result, and produces a new reading, and so on.

Figure 3. Digital filter modes affect reading speed. The repeating filter takes a new block of A/D conversions for each reading result in a); the moving filter computes a rolling average for each reading result in b).

Many DMMs have an auto-zero function that’s intended to compensate for input amplifier offset drifts. During an auto-zero cycle, a classic dual-slope integrating A/D converter switches through five distinct periods (Figure 4):

Figure 4. With auto-zero on, the DMM’s A/D converter runs twice as many cycles as with auto-zero off.

If you disable auto-zeroing, the A/D conversion cycle excludes zero integration and its reference integration stage, significantly shortening the overall conversion period and increasing measurement speed. But if the input amplifier drifts, the reading result will drift proportionally. This drift may be critically important at, say, the 100 mV level, but unimportant at the 10 V level. Therefore, you’ll need to make some judgements about when to use auto-zero functions.

Trigger delay is the period between receiving a trigger signal and starting the A/D conversion (Figure 5). Trigger delay is useful when signals have significant settling times, such as in multi-channel test systems where relays make connections to various inputs. Starting the A/D conversion before the signal settles properly will result in noisy, inaccurate, and unrepeatable measurements.

Figure 5. Trigger delay lets you optimise signal settling times.

Updating the instrument’s front panel display takes time, particularly with information-intensive displays. Displays that deliver multiple measurement parameters typically take up to 5% more time to update than simple alphanumeric displays.

One very effective way to speed measurements is to hold a number of results in the DMM’s local memory before transferring them as a block to the system controller. This technique avoids the repetitive latency that accumulates from requesting, arbitrating, and controlling bus systems such as GPIB.

Calculate Final System Throughput

The system’s total measurement speed is the sum of all its instrument configuration parameters plus all its switch configuration parameters. Although a fast DMM potentially allows blazing speed, the way you configure it can reduce your measurement throughput very considerably.

As an example, consider a test set-up that sequentially switches and measures six DC voltages, using a Keithley Model 2001 DMM and Model 7001 switch unit. The switch unit takes about 9.1 ms for each switch/read cycle (3 ms settling time, plus 6 ms actuation time, plus 0.1 ms trigger actuation time). Meanwhile, the DMM can take up to 2k readings/s, but you’ll optimise its read-rate/accuracy trade-offs to suit the overall system constraints.

On the DMM’s 1 VDC range, you select 1 PLC signal integration time, filter off, auto-zero off, delay off, and display on, and the DMM takes about 24 ms to produce each result. Together, the switch and DMM yield a total measurement time of 33.1 ms/ switch/read, or about 30 switch/ read cycles/s.

By carefully studying all the parameters of the elements within your system, and considering the data flow throughout, you can transfer the approach outlined here to virtually any instrument or ATE system, and maximise your test system’s throughput.