Europium thenoyltrifluoroacetonate (EuTFC) has an optical luminescence line at 612 nm, whose activation efficiency decreases strongly with temperature. If a sample coated with a thin film of this material is micro-imaged, the 612 nm luminescent response intensity may be converted into a direct map of sample surface temperature.
Micro-electronic devices often undergo significant self-heating when biased to their typical operating conditions. This paper describes a convenient optical micro-imaging technique which can be used to map and quantify such behavior. Europium thenoyltrifluoroacetonate (EuTFC) has a 612 nm luminescence line whose activation efficiency drops strongly with increasing temperature, due to T-dependent interactions between the Eu3+ ion and the organic chelating compound. This material may be readily coated on to a sample surface by thermal sublimation in vacuum. When the coating is excited with ultraviolet light (337 nm) an optical micro-image of the 612 nm luminescent response can be converted directly into a map of the sample surface temperature. This technique offers spatial resolution limited only by the microscope optics (about 1 micron) and time resolution limited by the speed of the camera employed. It offers the additional advantages of only requiring comparatively simple and non-specialized equipment, and giving a quantitative probe of sample temperature.
Many electronic devices undergo strong self-heating when electrically biased to their normal operating conditions. This is usually due to a combination of low thermal conductivity (such as in semiconductors) and high power dissipation density. Furthermore, in devices with a semiconducting-like electrical resistivity (i.e. with ∂ρ/∂T < 0) it has long been known that there exists the possibility of localized thermal runaway under certain biasing conditions 1,2, in which the bias current flows not uniformly through the device, but rather in narrow filaments which are associated with highly localized self-heating, typically on a scale of microns.
Understanding such self-heating physics may in some cases be essential for optimizing the design of a particular device, meaning that techniques for imaging temperature on micron scales are very useful. There has been a recent resurgence of interest in such techniques from two areas of technology development. The first of these is for imaging quench processes in high-temperature superconducting tapes in which thermal micro-imaging allows quench nucleation sites to be identified and studied 3,4. The second application is for understanding self-heating in stacked intrinsic Josephson junction terahertz sources, which are fabricated from Bi2Sr2CaCu2O8. These have the combination of low thermal conductivity and semiconductor-like electrical conductivity along the relevant direction of current flow (i.e. their crystalline c-axis) described above. Not only do they experimentally show complex inhomogeneous self-heating behavior 5,6,7,8,9,10,11 it has been theoretically predicted that this may be beneficial for THz power emission 12,13.
A number of techniques exist for imaging the temperature of a sample at microscopic length scales. The thermoluminescent technique described here was originally employed for semiconducting devices near room temperature 14,15,16 but has more recently been applied at cryogenic bath temperatures to the superconducting tapes and THz sources described above 3,4,10,11. Improvements in the resolution and signal-to-noise performance of CCD cameras have enabled considerable performance improvements in this technique over the last few decades. The Eu-coordination complex europium thenoyltrifluoroacetonate (EuTFC) has an optical luminescence which is strongly temperature dependent. The organic ligands in this complex effectively absorb UV light in a broad band around 345 nm. The energy is transferred radiation-less via intra-molecular excitations to the Eu3+ ion, which returns the complex to its ground state through the emission of a luminescence photon at 612 nm. The strong temperature dependence arises from the energy transfer process 17 making for a sensitive thermal probe of an object coated with this material. When the coating is excited with a near-ultraviolet source — such as an Hg short-arc lamp — regions with lower luminescence intensity correspond to higher local temperature. The resulting images are limited in spatial resolution by the resolution of the microscope optics and the wavelength of the luminescence (in practice, to around 1 micron). Depending on the signal-to-noise ratio required, time resolution is limited only by the shutter speed of the camera, and more fundamentally by the decay time of the luminescence (no more than 500 μs) 15. These characteristics make the technique a very fast probe of device temperature, which yields direct temperature measurements, using comparatively simple and economical equipment.
Variations of this technique published in the past by other groups have employed small concentrations of Eu-chelates dissolved in polymer films and spin-coated on to the sample surface 3,4. This results in a coating which is highly uniform locally, but which has significant thickness variations at steps in the sample topography — such as commonly occur in microdevices — resulting in strong spatial variations in the luminescent response which can give artifacts in the images. The technique variation which we describe here employs thermal sublimation in vacuum. Not only does this avoid the macroscopic film thickness variation problem, but the higher EuTFC concentration achieved per unit area significantly improves the sensitivity and reduces the image acquisition time. A related technique employs a coating of SiC granules on the surface instead of the EuTFC 7,8,9. SiC offers temperature sensitivity comparable to the EuTFC coatings described here, but the size of the granules limits the smoothness and resolution of the resulting images.
Several other techniques exist, which offer different combinations of advantages and disadvantages. Direct infrared imaging of blackbody radiation from the sample is simple and has spatial resolution of a few microns, but is only effective when the sample is significantly above room temperature. Scanning probe thermal microscopy techniques (such as scanning thermocouple microscopy or Kelvin probe microscopy) offer excellent sensitivity and spatial resolution, but have slow image acquisition times, necessarily limited by the scanning speed of the tip, as well as requiring highly complex equipment. Scanning laser or scanning electron beam thermal microscopy measures the voltage perturbation when a modulated beam is rastered across the surface of a current-biased device 6,7,18. This offers excellent sensitivity, and is somewhat faster than scanning probe techniques, but once again requires highly complex equipment, and also gives an indirect, qualitative map of the sample temperature.
1. Preparation of Sample for Coating
NOTE: If possible, remove all organic contamination from the surface of the sample to be thermally imaged. Any such contamination may react with the deposited EuTFC film and alter its luminescent response, causing position-dependent artifacts in the resulting thermal images. This is of particular importance with samples with Au surface electrodes, which tend to attract organic contamination from the atmosphere. Remove any particles or dust sitting on the sample surface at the same time, since these may result in artifacts also. The authors recommend the following procedure:
2. Preparation of Coating System for EuTFC Deposition
3. Deposition of EuTFC Thin Film by Thermal Sublimation
4. Installation of Sample in Measurement Cryostat
5. Collection of Thermal Image Data
6. Calibration of Results
7. Sample Storage and Film Re-use
An example of a typical measurement configuration for conducting this experiment at cryogenic bath temperatures is shown in Figure 1a, while a typical curve of 612 nm luminescent response intensity versus temperature is plotted in Figure 1b.
Figure 2 shows an example of typical thermal images of self heating in a Bi2Sr2CaCu2O8 THz source, which consists of a 'mesa' of stacked 'intrinsic' Josephson junctions with dimensions 300 x 60 x 0.83 microns, fabricated on the surface of a single crystal, and having a superconducting Tc of 86 K.
In such a device, the current flow is along the c-axis direction (i.e. into the plane of the page as shown in the images) due to the extremely anisotropic electrical resistivity of this material. As shown in Figure 2a, ρc(T) for Bi2Sr2CaCu2O8 falls strongly with increasing temperature, allowing the possibility of thermal instabilities and localized thermal runaway under certain biasing conditions. Thermal images of the device are shown in Figure 2d, which were collected as described in the text under 160X magnification, using summed exposures of 4 x 2 s on a 1,024 x 1,024 pixel CCD camera with 16-bit resolution, Peltier-cooled to -50 °C. The sample was illuminated with a short-arc Hg lamp using a 500 nm short-pass filter, and net intensity of approximately 1 W/cm2. To avoid the requirement of normalizing the images by an un-self-heated area as described in section 5.6, the lamp was operated using a variable iris with closed-loop feedback to keep illumination intensity constant over time.
The images reveal a localized hotspot, where local self-heating gives rise to self-sustaining filament of current flowing through the device in the c-axis direction. In this filament, the current density is over 5 times higher than in the rest of the mesa. The current-voltage characteristic for the mesa at Tbath = 25 K is shown in Figure 2b. This contains hysteretic jumps associated with the nucleation/annihilation of the hotspot at around Ibias = 11 mA, and with the jumping of the hotspot from the electrode end of the mesa to the opposite end between 40 and 60 mA. Figure 2c shows longitudinal cross-sections of the mesa surface temperature under different bias conditions. For the camera and imaging conditions used here, the temperature noise is around 0.2 K, when smoothed over a diameter of 4 microns, corresponding to a 5 x 5 pixel region at this magnification. The lines visible in Figure 2d at the edges of the mesa and of the electrode are artifacts due to reflection off near-vertical sidewall surfaces.
Figure 3 shows raw image examples of situations which should be avoided as described in the protocol. Figure 3a shows a 612 nm luminescent image in which the film was sublimated using EuTFC in which mm-sized lumps were present. (See step 2.4.) These sublimated violently when heated, depositing particles of EuTFC several microns in diameter on to the sample. Figure 3b shows a sample whose EuTFC coating has crystallized into domains after 16 hours at 150 K, resulting in uneven and noisy luminescent response. (See step 4.6.)
Figure 1: Thermal Imaging setup and typical calibration curve. (a) Configuration of Microscope, UV light source, and cryostat with optical window, modified from reference 10. (b) Response curve normalized to 10 K for 200 nm sublimated EuTFC film.
Figure 2: Bi2Sr2CaCu2O8 mesa THz source: I-V characteristics & thermal images. (a) (Main) Plot of device resistance against temperature. Blue squares plotted below Tc are values extrapolated from I-V curves shown in inset. (b) I-V characteristic showing hysteretic switching of Josephson junctions in device at Tbath = 25 K, for current-biased mesa. Insets (i) and (ii) show jumps in mesa resistance associated with hotspot nucleation and relocation respectively. (c) Longitudinal temperature cross-sections of mesa. (d) Thermal images at Tbath = 25 K, modified from reference 11, with conventional optical micrograph of mesa shown at left. Please click here to view a larger version of this figure.
Figure 3: Examples of problems to avoid with EuTFC film. (a) Film sublimated without removing large crystallized lumps from EuTFC powder, resulting in lumps deposited on sample. (b) Film (deposited on a different mesa) which has undergone local crystallization after 16 hours in cryostat at 150 K, showing uneven luminescent response. Please click here to view a larger version of this figure.
As demonstrated by our results, the technique described in this article yields high-resolution thermal images of microdevices, with good sensitivity and using only simple optical microscopy equipment. The advantages of this technique relative to alternative methods (which will be discussed below) are strongest at approximately 250 K and below, meaning that its most important applications are for studying the self-heating of devices which are designed to operate at cryogenic bath temperatures. These include superconducting current tapes (where quench nucleation is of key engineering interest), narrow band-gap semiconductors for optical detection, and novel high-Tc electronic devices whose resistance drops with increasing T.
If the technique is to work with optimum sensitivity, then it is critical to follow correct procedures for the deposition of the film. The sample surface must be cleaned thoroughly (protocol steps 1.1 to 1.5), the EuTFC powder must be carefully ground to remove any lumps which can adversely affect the uniformity of the film (step 2.4), and the film sublimation must occur at the correct rate in order to preserve the correct chelation of the Eu3+ ion (steps 3.3 and 3.4). Recrystallization of the film at cryogenic temperatures may increase the experimental noise level, but this problem can be reversed as described in step 4.7. The illumination and exposure parameters which should be used, and the resulting signal-to-noise, depend on the requirements of the experiment. Here we discuss some of the considerations which limit the performance of the technique.
There are four main possible contributions to the noise in this experiment, namely photon shot noise, microscopic variation in the luminescent response of the film, variations in camera pixel sensitivity, and camera dark count shot noise. Where I is the excitation illuminance (in incident photons per unit of pixel-equivalent sample area), F(T) is the T-dependent overall luminescent conversion efficiency for each pixel-equivalent area of the film (which is affected by the local film thickness), S is the CCD count yield from a pixel per incident photon (at = 612 nm), and D is the number of dark counts collected over exposure time t, then when averaged over P pixels, these parameters will be approximately normally distributed as follows:
σF(T) depends on the uniformity of the EuTFC coating, while the standard deviation σS in pixel-to-pixel light sensitivity and dark count rate standard deviation σD depend on the performance of the camera. The counts collected over P pixels for time t therefore have mean:
where the last term corresponds to the dark count contribution, and variance:
Therefore the standard error in the measured temperature when averaged over P pixels with total exposure time t is given by:
For a highly uniform film and a CCD with low pixel response non-uniformity , the terms in σF(T) and σS respectively may usually be neglected. The temperature error thus simplifies to:
For the conditions normally employed in this technique, the rate of luminescent photon collection is of the order of 5000 photons per pixel per second. For a modern cooled CCD camera, the rate of dark counts and thus σD is significantly less than this, meaning that σT is usually limited by photon shot noise 19. If σD can be neglected, then the temperature error simplifies further to:
Increasing the illumination intensity thus reduces the exposure time required for any given σT, especially in exceptional cases where the luminescent yield is low (e.g. at temperatures close to 300 K), and where dark counts are in fact significant. However, intense UV illumination may photodope carriers into semiconducting samples, and break Cooper pairs in superconducting ones, thereby perturbing the properties of the device being studied. In samples whose surfaces have a weak thermal path to the cold bath, strong illumination may also introduce a heat load which causes a significant rise in the sample temperature.
All of these considerations may sometimes necessitate low illumination intensities and longer exposure times. As a modification, shorter exposures may be required to image fast phenomena such as current filament oscillation or breathing modes 20, or the millisecond timescales of quench development in superconductors. Where high signal-to-noise ratios in absolute temperature measurements are required, then longer total exposure times are called for. This may require summation of multiple exposures, depending on the bit resolution of the CCD electronics. Image-intensified cameras have close to single-photon detection efficiency, and offer a more attractive trade-off between image noise, illumination intensity, averaging area, and exposure speed, albeit at higher system cost.
In summary, the thermoluminescent imaging technique which we describe here offers a direct quantitative measure of sample surface temperature, with high temporal and spatial resolution. It is also effective at a wide range of temperatures, from 5 K to over 300 K. As described in the Introduction, alternative techniques exist, but each of these offers a combination of advantages and disadvantages.
Scanning probe techniques offer excellent sensitivity, at the cost of long measurement times and highly specialized equipment. A recently-published pyro-magneto-optical technique also offers excellent sensitivity 21. However, this technique relies on a ferrimagnetic garnet indicator crystal placed on top of the sample, which limits spatial resolution, especially where the sample is not topographically flat. At temperatures above 300 K, the luminescent yield from EuTFC becomes low, and direct imaging of infrared blackbody radiation from the sample becomes a more effective technique.
The authors have nothing to disclose.
Work at Argonne National Laboratory was funded by the Department of Energy, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357, which also funds Argonne’s Center for Nanoscale Materials (CNM) where the patterning of the BSCCO mesa was performed. We thank R. Divan and L. Ocola for their help with sample fabrication.
Europium thenoyltrifluoroacetonate powder | Sigma-Aldrich | 176494-1G | Also known as Europium tris[3-(trifluoromethylhydroxymethylene)-(+)-camphorate] |
Mercury short-arc lamp with flexible light guide | Lumen Dynamics | X-Cite Exacte | Light source includes internal iris and photosensor for output intensity feedback. |
Peltier-cooled CCD camera | Princeton Instruments | PIXIS 1024 | 1024 x 1024 pixels, 16-bit resolution |
610 nm band-pass filter | Edmund Optics | 65-164 | Passband has CWL 610 nm, FWHM 10 nm |
500 nm short-pass filter | Edmund Optics | 84-706 | OD4 in stopband |
Helium flow cryostat with optical window | Oxford Instruments | MicrostatHe2 | |
high vacuum grease | Dow Corning | ||
Digital Current source | Keithley | Model 2400 | Computer-controllable current & voltage source |
Digital Voltmeter | Hewlett-Packard | Model 34420A | Digital Nanovoltmeter now available as Agilent Model 34420A |