DLTS User Meeting 2022 – Q&A

July 25, 2022 by Meng Li
DLTS User Meeting 2022

Following the success of the deep-level transient spectroscopy (DLTS) User Meeting last year, we were delighted to host the second User Meeting in 2022. As a powerful and commonly used technique to investigate defect concentration and trap level in semiconductor devices, DLTS has drawn more and more attention, not only from semiconductor physicists but also from other related fields. 

To continue contributing to DLTS and similar applications, this User Meeting intended to bring together the community who may face common measurement challenges and allow them to share core competencies and know-how. Once again, we thank all the speakers and attendees.

In this year's DLTS User Meeting, we focused on:

  • Celebrating the success of DLTS researchers at different academic levels;
  • Informing the community of the different types of DLTS and related techniques;
  • Fostering relations and exchanging ideas via open Q&A sessions and a virtual coffee break; and
  • Empowering researchers by sharing the best tips and tricks, in both hardware (setup) and software (API). 

If you missed the User Meeting, recordings of the talks and tutorials can be viewed here.

This blog post presents a selection of questions raised during the meeting, and were answered live or retrospectively by Xiaobo Hu (XH), Dino Klotz (DK), Sandhya Tammireddy (ST) or me (Meng Li, ML). If you have any follow-up questions, please get in touch with us.

Q&A

Does a given type of defect have a definite activation energy and time constant for migration?

Answer from ST: It is important to note that atomic defects in semiconductors create additional electronic defects with states in the bandgap and these atomic defects are mobile in perovskites. It seems that in perovskites, we measure the mobile atomic defects rather than the electronic defects created by them. When an electric field is applied to perovskite material, in capacitance frequency spectra, the resulting time constant i.e., the rate at which an ionic defect performs the jump between one lattice site to another depends on the neighboring bonds and location of the defect. These two quantities are specific for an ionic defect and their summation is the activation energy that can be accessed when the time constants are measured as a function of temperature.

An electrode polarization will have an intensive influence on the IS results (especially in a very low-frequency range). Could you please let me know how this effect is cut from your measurement data (if a such technique is used to reduce the EP effect)?

Answer from ST: Indeed. We also observe the influence of electrode polarization in our low-frequency data. This effect is always part of the IS signal. We are currently working on methods such as Modulus spectroscopy where low-frequency defects can be evaluated with improved resolution and without the influence of resistances.

How do you distinguish the deep-level traps/defects and shallow-level traps/defects in your work?

Answer from ST: It must be noted that we observe ionic defects in perovskites, not electronic defects. The categorization of a defect being shallow or deep only applies to electronic defects with respect to where they are located in the band gap. It is possible to distinguish between shallow and deep “electronic” defects based on their energy levels in the bandgap. Slopes in the Arrhenius plot of emission rates would lead to the activation energy of an electronic defect. If the obtained activation energies are comparable to thermal energy then the measured electronic defect can be referred to as a shallow defect otherwise a deep defect. In contrast, this is not the case for “ionic defects” since the activation energy of an ionic defect is unrelated to the bandgap but only implies the energy required to create and move an ionic defect. However, Ionic defects can also create electronic defects with states in the bandgap.

You mentioned point, line, and grain boundary defects, 0D, 1D, and 2D respectively. Are there any ways to estimate the density of these, separately?

Answer from ST: We only observe point defects (0D) i.e., mobile ionic defects, and the dominant ionic defect seems to migrate at the boundaries. It’s possible that 1D and 2D defects are less mobile. We can’t say for sure.

What are the solar cell structures? What is the likelihood of measuring defects from the charge transport layers?

Answer from ST: The device architectures can be found in (https://doi.org/10.1021/acsenergylett.1c02179). It is indeed possible that the transport layers would have an impact on defect characterization. However, when we compared our defect (β) migration rates to the rates or conductivities reported in literature where metal/perovskite/metal architecture was used, both coincided well. This implies that the high-frequency defect likely originated from perovskite rather than transport layers. We recommend doing such comparisons in the Arrhenius representation.

Since you get similar defect signatures in all perovskites, it is possible due to charge transport layers. Did you try to study with and without charge transport layers?

Answer from ST: We compared migration/emission rates or conductivities of the defect observed in three perovskite solar cells to the conductivities reported in literature where metal/perovskite/metal structures were used. All these values coincided well, indicating that the observed defect (β) is originated from perovskite rather than the transport layer. We are currently working on Au/perovskite/Au structure and the measured spectra show the same migration rates for defect β.

At which AC frequency do you measure, and why?

Answer from ST: DLTS transients were measured at 80 kHz. The frequency is chosen in a way that it is large enough for the capacitance signal to reach saturation and small enough to avoid the influence of series resistance. A good practice would be first to measure capacitance as a function of frequency and select the minimum frequency at which the signal reaches a constant value.

Could you please comment more on why you assumed the detected defect as mobile ions rather than immobile?

Answer from ST: Only moving charges will contribute to a change in capacitance in the impedance spectra. If there are no moving species in the film we should only observe a constant capacitance that is originated due to the geometry of the device. Since we see two different capacitance steps in our measurements, the signal is originated from a mobile species.

Where does the emission rate formula come from? What assumptions are made in its formulation?

Answer from XH: We assume the emission rate (inverse carrier time) is limited by Schottky-Read-Hall (SRH) recombination and further follows thermionic emission (Boltzmann distribution) in temperature. This is valid in many cases. Check Sze's book <semiconductor devices physics and technology> for details.

For the temperature control loop, do you use the temperature controller or set up the loop in your Python script?

Answer from XH: We did it manually. 

Can you share how to design a test fixture to solve the problem of electrostatic shielding during testing?

Answer from XH: We wrapped Aluminium foil around the setup to reduce radiation noise.

Do you also use an optical DLTS technique for the results which can be found in your papers?

Answer from XH: Not yet.

How did you select the frequency 100 kHz? Have you tried lower frequencies such as 10 kHz or 1 kHz or less? Can you please explain?

Answer from XH: This depends on the equivalent circuit model of the device-under-test (DUT). The frequency is not so important as long as the DUT behaves mainly capacitively. Please check the presentation from Roman Schifano in last year's meeting for a more detailed answer.

How the carrier lifetime is measured? Was it a separate test apart from DLTS?

Answer from XH: Carrier lifetime is derived and is inversely related to the emission rate. 

What is the minimum trap density you can detect?

Answer from XH: About 5e14 cm-3.

Could we measure 1/f noise signal based on this instrument?

Answer from DK: Yes, the Sweeper Module of LabOne has a noise density spectrum sweep mode, available in the sweeper advanced settings. This can measure the whole noise of the setup, including the 1/f noise.

Can we do similar experiments using MFLI or HF2LI?

Answer from DK: It is possible, but as MFLI and HF2LI measure only current (and voltage), you will need to calculate the capacitance in post-processing with proper circuit models. The capacitance transient is not available in real-time. Adding the MF-IA option to the MFLI would allow for capacitance transients to be measured

What causes the delay between the bias pulse to the LED and the light detected by the photodetector? Is it fast enough or can it be sped up?

Answer from DK: The LED driver does not provide information about the delay between the input signal and the output signal, and there will be a delay of several microseconds due to the electronics' response time. It is always recommended to correlate the obtained results with the actual output signal, such as the light, instead of just the trigger signal.

For the first technique, you send a light pulse, for the IMPS technique you send a sine wave. Are there options for other signals?

Answer from DK: The MFIA can generate common waveforms such as square, sawtooth, triangle, and sine from its Auxiliary Outputs. However, if you want to have better control over these, you may want to use a high-end external arbitrary function generator, and feed it into the Auxiliary Input of the MFIA Impedance Analyzer. This perturbing function can then be added to the ac test signal that is used to measure the sample impedance.

What is a preference of DLOS over the standard DLTS? And what about current-mode DLTS (I-DLTS)?

Answer from DK: The techniques Optical-DLTS (ODLTS) and current-mode DLTS (I-DLTS) differ from conventional DLTS only by how the defects are charged. In all three techniques, the defects are thermally stimulated, which limits DLTS to defect states not deeper than 1 eV. DLOS bypasses the problem by using monochromatic light that successively covers all energies within the bandgap. The defects measured by DLOS are optically stimulated.

Many thanks to several audiences who helped to discuss and to answer it, especially Piotr Kruszewski from the Laboratory of Semiconductor Characterization at the Polish Academy of Sciences.

Can the APIs reinitialize the MFIA to its default state (overwriting the existing settings of all nodes)? Is this already included when an interface is opened? If not, is there any other way of assuring the instrument is in a known state before controlling it?

Answer from ML: Rebooting the instrument is the easiest solution. In addition to that, with the API, you can also call 'disable everything' to roll back to the default settings, or just load an existing settings file to change to other preferred settings.

Can the HF2LI use the same Python code as you presented in the webinar?

Answer from ML: We can follow the same protocol to copy and paste the syntax from the LabOne command log. But the detailed node structure is different. Particularly, the HF2LI only gives voltage and current, and you may need to calculate the capacitance manually.

Your talk was about Python - what about other APIs/languages you can use to interface the MFIA?

Answer from ML: In addition to Python, MATLAB, LabVIEW, C, and .Net are also supported.