Any characterisation lab engineer responsible for developing new materials, devices, or processes knows that, just like dc current-voltage (I-V) and capacitance-voltage (C-V) measurements, performing ultra-fast I-V measurements has become essential. Making these measurements involves generating high-speed pulsed waveforms and then measuring the resulting signals before the device under test can relax.
Early implementations of high-speed I-V testing, often called pulsed I-V test systems, were developed to address applications such as characterizing high-k dielectrics and silicon-on-insulator (SOI) isothermal testing. They also were used to generate the short pulses needed to characterise flash-memory devices.
Pulsed I-V measurement techniques are necessary because the insulating substrates of traditional dc I-V techniques will induce SOI devices to retain the heat self-generated by the test signal, skewing their measured characteristics. Pulsed test signals minimise this effect.
In the past, high-speed-pulse/measure test systems typically consisted of a pulse generator, a multichannel oscilloscope, interconnect hardware, and software to integrate and control the instruments. Unfortunately, latencies plagued these systems, complicating the coordination of signal source and measurement functions. Depending on the quality of the instruments and how well they were integrated, this approach could also limit the length of the pulses and their duty cycle.
Despite these limitations, users of these early pulsed I-V test systems began looking for ways to apply them to a variety of other characterisation tasks, including nonvolatile memory testing, ultra-fast negative bias temperature instability (NBTI) reliability testing, and many other applications. Nonetheless, the somewhat restricted dynamic range of these systems relegated them to more or less a specialty technology.
To become a mainstream test technology, the next generation of ultra-fast I-V testing systems needed to provide a very broad source and measure dynamic range. That meant the ability to source sufficient voltage to characterise flash-memory devices, as well as voltages low enough to handle the latest CMOS processes.
For example, consider an embedded flash device in a CMOS process. The flash device might require up to 20 V to program, but the CMOS process is running on 3 V. Therefore, the test system must be able to supply voltages for both requirements.
The next generation also required a broad enough current range to handle the newest technologies, as well as rise times and pulse widths that could meet the needs of a wide range of applications. It had to be simple to use and incorporate an interconnect system that would allow the system to deliver accurate results reliably.
Today, parametric analysers integrate ultra-fast I-V source and measurement capabilities for characterising a growing range of device characteristics, particularly NBTI and positive bias temperature instability (PBTI) degradation. By allowing researchers to make quick and consistent device reliability measurements, ultra-fast I-V measurement tools improve the accuracy of designed-in reliability (DIR) lifetime measurements, which support modeling for device and circuit design.
Continue on next page
Until recently, some researchers were forced to configure their own ultra-fast bias temperature instability (BTI) test systems. These in-house-developed systems typically combine a pulse generator or arbitrary waveform generator (AWG) with an oscilloscope featuring current probes or some type of transimpedance amplifier to help measure low current.
Although it’s possible to build a BTI system that suits a very specific set of electrical conditions after careful selection of the instruments and interconnect, several major technical challenges remain:
• Waveform generation: Standard pulse generators and AWGs are designed to generate a waveform on a fixed recurring interval, rather than the Log (time) scale required for most reliability tests, including NBTI and PBTI testing.
• Measurement timing and data storage: Although oscilloscopes can be configured to trigger based on a waveform feature (such as a falling edge), they aren’t designed to store samples selectively for specific portions of the waveform. As a result, very large data sets must be stored for post-processing. Only the most expensive oscilloscopes or those with costly memory-expansion options can store enough data to compensate for these shortcomings.
• Precision, accuracy, and sensitivity: BTI is a highly dynamic phenomenon that requires sensitive, high-speed measurements for accurate characterisation. Assuming all other factors are constant, measurement physics largely define the relationship between measurement speed and sensitivity. When making sub-millisecond measurements, all sources of noise must be considered. For sub-microsecond applications, even quantum effects can’t be ignored.
Oscilloscopes, current probes, and transimpedance amplifiers all have independently defined performance specifications, and they aren’t necessarily optimised to work together. It’s often very difficult to combine these components to attain optimal performance across a wide dynamic range, with the ultimate goal of achieving precision accurate measurements at high speeds.
• Interconnect: Systems built in-house typically use splitters and bias tees, which limit the performance of the test setup. For example, a bias tee might limit bandwidth from 100 ns to 10 µs. Although this works for high-speed measurements, it prevents any meaningful pre-stress and post-stress dc measurements as part of the stress-measure sequence. It also prevents measurements in the intermediate timing range of 10 µs to dc.
• Test control and data management: Traditional oscilloscopes don’t support data streaming, so the transfer of results must wait until the test ends. Once the test is complete, massive amounts of data need to be transferred to the control computer for post processing. This requires the parsing of complex waveforms into individual test results, followed by further reduction of the data into actual measurements.
• Test termination: Since test results can’t be analysed until the data is transferred from the oscilloscope, the test duration must be determined prior to test initiation. This makes it impossible to terminate the test based on parametric shifts or detect catastrophic failures in real time.
• Automation: Wafer-level or cassette-level automation requires control of both the test instruments and the wafer probe station, which usually isn’t possible with in-house systems. Also, the incorporation of sophisticated features like conditional test termination would add complexity to the custom software necessary to run such a system.
Continue on next page
• Larger channel count: Even if an in-house system works well when initially installed, system integrators may need to increase the channel or test system count to address evolving applications. Such a task can be extremely complicated when attempting to upgrade a custom system. In addition, typical test system maintenance issues (e.g., calibration, operation, and correlation of these custom setups) may require a disproportionately high amount of technical resources, which are often in limited supply.
Eliminating The Shortfalls
The latest generation of parameter analysers can be configured to minimise or eliminate many of the shortfalls associated with in-house BTI characterization systems. Rather than a separate pulse or waveform generator and oscilloscope, they now combine these functions in high-speed source and measure modules that allow for tight timing coordination.
These modules are fully integrated with the parameter analyser, enabling them to take advantage of the system’s data-storage and automation capabilities. Chassis-based systems also make it easier to increase the number of high-speed channels—just add more modules.
Thanks to the latest generation of parameter analysers, ultra-fast I-V, dc I-V, and C-V measurements can be integrated into the same test sequence. This is a particularly valuable capability for the growing number of applications that use multiple measurement types.
One such example is charge pumping (CP), which typically requires pulsing a gate voltage and measuring a dc substrate current simultaneously. Another involves determining the electrical characteristics of photovoltaic (solar) cells, which means measuring current and capacitance as a function of an applied dc voltage.
Keithley’s Model 4200-SCS semiconductor (Fig. 1) characterization system has long supported precision dc I-V measurements and C-V measurements (with an optional C-V module). Also, the recent introduction of the Model 4225-PMU Ultra-Fast I-V Module and Model 4225-RPM remote amplifier/switch makes it possible to add ultra-high-speed sourcing and measurement to create a system that’s also optimised for emerging lab applications.
Among these emerging applications are ultra-fast general-purpose I-V measurements; pulsed I-V and transient I-V measurements; flash, PCRAM, and other nonvolatile memory tests; isothermal testing of medium-sized power devices; materials testing for scaled CMOS, such as high-k dielectrics; and NBTI/PBTI reliability tests. Figure 2 shows many of these applications for the Model 4200’s dc I-V and ultra-fast I-V sourcing and measurement ranges.
Ultra-fast I-V sourcing and measurement pick up where more traditional dc I-V (Fig. 2, again) capabilities leave off. Note how traditional source measure unit (SMU) designs can source and measure currents up to about 1 A, and down to about a picoamp. Although the addition of a remote preamplifier allows designers to resolve signals as low as 0.1 fA, these dc-I-V-only configurations’ best speed was about 10 ms.
In contrast, measurements as fast as 10 ns are possible with ultra-fast I-V solutions, which is critical for applications that involve characterising device recovery. Optional remote amplifiers designed specifically for ultra-fast I-V testing extend the current resolution of these new solutions down to tens of picoamps.
Single-chassis systems that combine ultra-fast I-V source and measure instrumentation with remote amplifiers support a broader array of characterisation applications than ever before. The growing list of apps includes testing of phase-change memory devices, single-pulse charge trapping/high-k dielectric test, characterization of LDMOS or gallium-arsenide (GaAs) medium-power amplifier devices, SOI isothermal test, ultra-fast NBTI test, charge-based capacitance measurement (CBCM), and micro-electromechanical systems (MEMS) capacitor test.
Four sweep types are available to support this array of applications (Fig. 3). Transient I-V sweeps continuously digitise voltage and/or current. Fast-pulsed I-V samples voltage and/or current after the pulse has settled. Filtered pulse generates a variable pulse voltage while a dc SMU measures the resulting current. And, pulse stress/dc measure pulses the voltage, which is followed by a dc SMU measurement.
In addition to these traditional sweep types, the Model 4225-PMU incorporates a full arbitrary-waveform-generation capability. It simplifies the creation, storage, and generation of waveforms made up of up to 2048 user-defined lines segments. Each segment can have a different duration, which helps increase waveform generation flexibility.
Lee Stauffer is senior staff technologist for Keithley Instruments’ Semiconductor Measurements Group.