Open Access
Add to BibSonomy
Abstract
The motivation behind time-resolved Raman spectroscopy for planetary surface exploration is (1) to provide comprehensive identification of minerals (nearly all rock-forming minerals and weathering products) and many organics of prime importance including fossilized carbonaceous materials; (2) to do so ensuring that it is possible to characterize even the most sensitive materials that would be altered by current state-of-the-art pulsed lasers (e.g., dark minerals, organics). These goals are accomplished here using a lightweight, high-speed (MHz) pulsed ($\lt {100}\,{\rm ps}$) Raman spectrometer based on a high-speed microchip laser combined with a single photon avalanche diode detector array. Using a Mars analog sample set and an automated grid sampling technique, we demonstrate consistent identification of major minerals and kerogen detection at $\sim\ge {1}\%$ by volume, without losses typically associated with the two biggest problems: fluorescence interference and sample damage. Despite improvements, we find that time-resolved Raman spectroscopy is still limited by the availability of a suitable laser and detector. As technology advances and such devices become available, we expect that this technique will hold an important place in Raman spectroscopy for both commercial and planetary science applications. We also discuss the utility of Raman point mapping for planetary science (e.g., in comparison with other common techniques like infrared reflectance spectroscopy) and conclude that the choice of technique must be planetary mission-specific; one must consider whether incurring the time to map single microscopic points is worthwhile, and how many points would be sufficient to gain the required information to characterize the surface.
© 2020 Optical Society of America
1. INTRODUCTION
A. Background
1. Overview
Raman spectroscopy has been gaining interest over the past several decades as an important technique in a wide variety of fields [1–3]. Recent advances, especially in the areas of miniaturized and robust high-performing lasers and detectors, have led to the development of more compact instruments for applications in space exploration that were previously unattainable [4–6]. This work builds off of our previous work [7], where we demonstrated the effectiveness of time-resolved Raman spectroscopy, using commercially available pulsed microchip lasers and custom-built single photon avalanche diode (SPAD) time-resolved detectors, for enhancing science return in natural geologic samples relevant to planetary science. Here we apply newly developed pulsed laser technology, high-speed microchip (HMC) lasers to improve the sensitivity of the technique for some of the most challenging samples, especially those containing organics, and to simplify the operation of the instrument in a way that is suitable for autonomous operation on a planetary surface. In the following sections, we discuss the motivation behind using HMC lasers, the implementation of the lasers in our custom Raman instrument, science results showing improved performance, and remaining technical challenges.
2. Quantitative Raman for Planetary Surface Exploration
We focus in this work on the application of Raman to in situ, nondestructive planetary surface characterization, for example, using a miniaturized instrument mounted on the arm of a rover or lander. This is a noncontact method requiring no sample preparation that can be used to produce quantitative surface maps that preserve important contextual information [8,9]. Maps are obtained by collecting a large number of Raman spectra measured over a grid within a small sampling area (typically millimeters to centimeters) coincident with a microscopic image.
For natural geologic materials, there are some complexities to consider [10]. For this method to be quantitative, it is required for the laser spot size interrogating the sample to be smaller than the grain size of the material being sampled such that each spectrum obtained represents a distinct constituent. As the laser spot size on the surface becomes arbitrarily large, quantification is lost due to the uncertainty related to an unknown laser absorption volume in an inhomogeneous sample. An alternate way to achieve quantification is to collect and prepare a sample into a homogeneous mixture for laser interrogation, e.g., as in the ExoMars Raman spectrometer [4]. We concentrate here instead on the development of a Raman system requiring no sample preparation, with a small spot size ($\sim\!\!{10}\;\unicode{x00B5}{\rm m}$ FWHM) that can sample rapidly and autonomously in a grid pattern over a chosen area coincident with a microscopic image (mm–cm scale) in order to provide a quantitative surface map of minerals and organics. The goal is a system that can perform this sampling without losses typically associated with phenomena like fluorescence and laser sample damage.
3. Fluorescence Interference
Fluorescence interference in Raman spectroscopy is widely cited as perhaps the greatest challenge across all fields in which Raman spectroscopy is applied. The problem is magnified by the fact that even trace amounts of impurities (e.g., rare-Earth elements and ${{\rm Mn}^{2+}}$ in minerals) and organics can produce strong fluorescence that interferes with the ability to detect Raman spectra. In the context of planetary science, the presence of fluorescence originating from organics is of prime concern for nearly all planetary bodies, and is expected to be highly variable throughout the solar system. It is known that organics such as polycyclic aromatic hydrocarbons (PAHs), compounds that are present throughout the solar system, can be highly fluorescent [11]. PAHs have been reported in carbonaceous chondrite meteorites, Martian meteorites, Stardust comet samples, interplanetary dust particles (IDPs), and interstellar matter. Their reliable detection and characterization is considered essential for determining the potential for habitability and life elsewhere in the solar system. On Earth, PAHs are products of the transformation of buried primary organic matter, forming geopolymers like kerogen as the result of biodegradation processes. A similar process is thought to have been possible on Mars, where potentially habitable surface environments are thought to have existed in the past. For example, minerals associated with ancient sedimentary deposits and aqueous alteration, such as phyllosilicates and sulfates, have been identified both from orbit and in situ [12–15]. Future missions will be aimed at seeking samples for return to Earth, likely from similar habitable environments that are most likely to contain evidence of past life [16,17]. Furthermore, many aqueously formed minerals from similar environments on Earth show high levels of background fluorescence [18–20]. In addition, fluorescence has been problematic in the characterization of Martian meteorites by Raman (e.g., nearly half of the 362 Raman spectra taken on the Zagami Mars meteorite at 532 nm yielded no Raman information due to fluorescence [21]). More recently, Cloutis et al. [22] reported on the interference of fluorescence in Raman spectra for serpentinites and related hydrated silicates of relevance to Mars; they conclude that while Raman is valuable in a large number of cases, reflectance spectroscopy is generally superior to Raman because it is not subject to fluorescence interference. In order to be relevant in anticipation of highly variable environments, there is a clear need for development of a Raman method that is not susceptible to fluorescence.
4. Time-Resolved Raman
For a complete description of the motivation behind time-resolved Raman, see Blacksberg et al. [7]. The primary advantages of this technique are (1) the use of a 532 nm excitation source (green) while eliminating interference from background fluorescence that can be severe with this excitation wavelength (2) operation under ambient light without the need for shielding, a significant advantage for field instruments, especially instruments designed for space exploration where in situ shielding can be complex.
Time-resolved Raman takes advantage of the differing time scales of Raman and fluorescence, collecting only the Raman while avoiding the fluorescence using a temporal gating technique. The ability to eliminate fluorescence using this technique is dependent upon the time scale of the specific fluorescence process and the temporal characteristics of both the laser and detector gate. Mineral fluorescence can be nanoseconds (ns) to milliseconds (ms) while organic fluorescence is typically shorter in the sub-nanosecond to nanosecond range [23,24]. Figure 1 shows how the parameters of the laser and detector gate affect the ability to eliminate background fluorescence.
Fig. 1. Illustration of time evolution of Raman and fluorescence shown for a 100 ps laser pulse. The green curve represents the Raman return. The blue, magenta, and red curves represent the return from fluorescence with different lifetimes, convolved with the laser pulse. The detector time gate is illustrated by the shaded area (1 ns gate duration with 550 ps rise time and 250 ps fall time). The detected signal is the product of the relative intensity and the detector sensitivity. It is clear that a significant fraction of the fluorescence return can be rejected, even when the lifetime is as short as the laser pulse duration. In order to maximize signal-to-noise ratio (SNR), the time gate is activated prior to the laser pulse. This allows most of the Raman signal to be collected while a significant portion of the fluorescence is rejected. As shown, gating is most effective when the fluorescence lifetime is longer than the laser pulse duration. The time gate duration is not important, but the detector fall time has a major impact on the short lifetime fluorescence rejection capability.
B. Approach to Overcoming Current Limitations
In order to understand the requirements for a Raman instrument that address planetary science questions, we must first look closely at how Raman would be performed remotely on another planet without supervision. Some basic guidelines are (1) we would need to collect a map of spectral point measurements, for example in a predefined grid pattern overlaid on the microscopic image of the area under study; (2) individual measurements would need to be performed quickly (seconds rather than minutes) in order to make the grid collection feasible within a typical mission operations timeline where there are strong constraints on individual instrument operation time, power, and data volume; (3) we would need to get usable data every time without the need to vary the laser power depending on the damage threshold of the sample, which would not be feasible to implement as it is most often done in the lab on a case-by-case basis.
These basic guidelines led us to identify two key improvements over our previous work needed to move forward. First, we need to improve fluorescence rejection, which means narrowing the laser pulse duration. Organics are often present in natural samples of interest for planetary science, and they can be dominated by sub-ns lifetime fluorescence obscuring weak Raman signatures that require more aggressive time gating. Second, we need to remove the need to vary the laser power for each sample point. We achieve this by lowering the energy per pulse to a level below the damage threshold. In turn, this requires a high repetition rate laser so that the average power on the sample is sufficient to perform fast measurements.
We aim to address these limitations in this work, and demonstrate that significant improvements can be made to the time-resolved Raman instrument. We aim to improve fluorescence lifetime rejection by reducing the effective time gate duration. The current limitation in this regard is the lack of availability of miniaturized and robust pulsed lasers with short pulse duration (${\rm target}\;\lt\;{100}\;{\rm ps}$). At the same time, we aim to improve sensitivity by lowering the peak laser power delivered to the sample, preventing sample damage, and allowing all measurements to be made at the same laser power, significantly simplifying operations. This in turn requires a higher repetition rate (target 1 MHz) in order to achieve a sufficient duty cycle and feasible spectral acquisition time (target 1 min per spectrum). The current technical limitation in this regard is the lack of availability of miniaturized and robust high repetition rate low pulse energy microchip lasers. The focus of this work is therefore on the development and demonstration of a new type of laser, the HMC laser, with short temporal pulse duration, and its integration into our custom time-resolved Raman spectrometer system for enhanced performance.
C. High Repetition Rate and Low Pulse Energy Lasers as a Solution
Recent developments in semiconductor saturable absorber technology have realized a significant leap in passively $Q$-switched diode-pumped solid-state (DPSS) microchip laser technology. These new lasers offer the same narrow spectral linewidth and excellent beam quality while operating at much higher repetition rates and at pulse energy levels appropriate for our application and with shorter pulse duration. They can deliver high average power with MHz repetition rate [25,26] to the sample while maintaining low pulse energy needed to avoid damaging sensitive samples. Second, they can deliver very short laser pulses ($\lt\! {30}\;{\rm ps}$) [27,28], thereby enabling fast time gating for effective rejection of short lifetime organic-related fluorescence. While the performance of these lasers is excellent for our application, they do present potential challenges in reliability that will be discussed.
2. EXPERIMENTAL
A. Instrument Architecture
1. High-Speed Microchip Laser
At the time of writing, no laser with the specifications needed (as discussed in Section 1.C) for the time-resolved Raman system was available commercially. Instead, the laser was built from the ground up, using a commercially available laser microchip (Batop GmbH, MCT-1064-100 ps). The laser microchip consist of a thin ${\rm Nd}:{{\rm YVO}_4}$ slab, bonded to a semiconductor saturable output coupler, acting as a passive $Q$ switch. When appropriately pumped with an 808 nm CW laser, the laser microchip emits 100 ps duration pulses with $\gt\!{20}\;{\rm nJ}$ energy at 1064 nm. The pulse repletion rate varies (up to $\sim\!{500}\;{\rm kHz}$) with the pump energy while the pulse duration and energy stay the same. As is typical for DPSS lasers, the linewidth is suitably narrow for Raman spectroscopy, $\lt\! {0.1}\;{\rm nm}$.
A schematic of the HMC laser is shown in Fig. 2, along with an image of the final packaged laser. Briefly, the pump laser is a wavelength-stabilized, single-mode (SM) fiber-coupled, diode laser (Lumics, LU0808M300). The fiber output is focused to produce a $\sim\!{40}\;\unicode{x00B5}{\rm m}$ FWHM Gaussian spot on the laser microchip. The optical fiber and lens are mounted at a small angle to eliminate back reflections. The temperature of the laser microchip is controlled using a thermoelectric cooler (TEC). The 1064 nm output from the microchip is collimated and sent through a miniature optical isolator (Fastpulse, 2120 ISO-2-1064) in order to eliminate any perturbing effects from backreflections. Following the optical isolator, the light is focused into a 10 mm long MgO:PPLN crystal (Covesion, MSHG1064-1.0-10), converting the optical pulses from 1064 nm to 532 nm with $\sim\!{60}\%$ conversion efficiency. Finally, the 532 nm light is collimated, and the remaining 1064 nm light is removed using a short-pass filter (Thorlabs FESH0700).
Fig. 2. Schematic of the HMC laser design (left) and photograph of the prototype packaged laser with the lid removed (right).
At this point, it is important to point out a significant limitation encountered with the current generation fast laser microchips: the limited lifetime of the saturable absorber output coupler. Failure of the saturable output coupler manifested as a sudden change in the laser output parameters (changed repetition rate, reduced stability, drop in power, etc.), rendering the laser useless for its intended purpose. After failure, each laser microchip can be “reset” by moving the pump spot to a new location on the chip. While this is useful for a laboratory demonstration, this issue, until solved, limits the applicability of these lasers outside the lab. In order to maximize the lifetime of the laser, strict temperature control of the laser microchip and elimination of backreflections were found helpful. However, the power of the pump laser was found to be the ultimate limitation. Thus, although we were able to run the laser at repetition rates up to 500 kHz, we chose to limit the laser repetition rate in this work to $\sim\!{150}\;{\rm kHz}$ in order to minimize the risk of laser failure during measurements. In this work, we observed a laser lifetime of a few hundred hours.
For reference, we used a conventional passively $Q$-switched 532 nm DPSS laser (TEEM Photonics SNG MicroChip), which delivers spectrally narrow ($\lt {0.1}\;{\rm nm}$) pulses with $\sim\! {1.5}\;\unicode{x00B5} {\rm J}$ pulse energy and $\sim\! {600}\;{\rm ps}$ pulse duration, at a repetition rate of 40 kHz. This is the same laser that was used in our previous work [7].
2. Overall Instrument Architecture
A schematic of the time-resolved Raman spectrometer used in this work is shown in Fig. 3. The instrument builds upon the work described in Ref. [7] but with some notable upgrades to accommodate the HMC laser.
Light from the HMC laser (described in the previous section) is coupled into a 10 m long SM fiber (Thorlabs, 460HP), which acts as the delay line, allowing the electronics sufficient time to synchronize the time gating with the arrival of the light to the SPAD detector. To facilitate the synchronization, a small portion of the light from the HMC laser is diverted to a fast triggering detector. The electrical signal from the triggering detector is sent to the detector electronics via a 10 ps resolution, computer controllable picosecond delay unit (Micro Photon Devices, Picosecond Delayer).
Light from the SM fiber is collimated and sent through a laser cleanup filter (Semrock, LL01-532-12.5) to remove any off-laser wavelength light generated in the fiber (e.g., Raman or fluorescence). The light is then delivered to the sample though a long-working distance 0.60 NA microscope objective (Qioptic, Optem ${20}\times$ M Plan APO LWD). The collimator-objective combination is chosen so that the diffraction limited spot at the sample is $\sim\!{10}\;\unicode{x00B5}{\rm m}$. The return light is collected by the same objective and routed through a dichroic edge filter (Semrock, LPD01-532RU-25) as well as an additional long-pass filter (Semrock, LP03-532RE-25) before it is coupled into an optical fiber, which sends the light to the miniature spectrometer. This short (1 m) optical fiber (Fiberguide multimode 50/125, 0.12 NA) was chosen to minimize temporal broadening and chromatic dispersion.
The miniature spectrometer (f/4, crossed Czerny–Turner design) is based on flight heritage spectrometers (e.g., LCROSS, LADEE, O/OREOS, and MSL [29], with minor modifications, built by Draper Laboratory. The spectral range is $-{50-2200}\;{{\rm cm}^{-1}}$, and the spectral resolution is $\sim\! {6}\;{{\rm cm}^{-1}}$.
The detector used in this work is a ${1024}\;\times\;{8}$ pixel SPAD array described in detail by Maruyama et al. [30]. This SPAD array was designed and fabricated using standard complementary metal-oxide semiconductor (CMOS) processes and features, 550 ps and 250 ps rise and fall times of the gate, respectively. In order to maximize signal-to-noise ratio (SNR), the spectrometer and SPAD detector array were aligned to collect a majority ($\sim\!{90}\%$) of the signal in the two center rows of the detector, effectively operating it as a ${1024}\;\times\;{2}$ array. Furthermore, in order to enable operation of the SPAD detector up to $\sim\!{1}\;{\rm MHz}$, the readout architecture was modified to minimize bottlenecks. Most notably, the readout field-programmable gate array (FPGA) firmware was used to accumulate the integrated counts per pixel over a large number of gates, before relaying this to the computer over a slower serial interface.
B. Calibration and Performance Limitations
Calibration of the instrument is a crucial topic to consider, and this instrument poses unique calibration challenges due to the short time scale temporal variation that is required to make it work. The wavenumber calibration can be performed exactly as in conventional Raman spectroscopy, i.e., by measuring and locating the peaks from a known sample (we use cyclohexane), but the relative intensity calibration, sometimes referred to as flat-field calibration, becomes significantly more complicated.
The purpose of the relative intensity calibration is to correct for pixel-to-pixel variations in, e.g., detection efficiency, which will show up in any measured spectrum as fixed pattern noise. The calibration is usually performed by measuring a calibrated broadband lamp or a calibrated fluorescent sample (such as NIST Standard Reference Material 2242). However, in this work, since the pulse we want to measure arrives during the fall time of the detector gate (cf. Fig. 1), minute pixel-to-pixel differences in fall time, or gate timing, will have to be accounted for. In order to do this, we use a supercontinuum laser (NKT Photonics, SuperK-Compact), with 20 kHz repetition rate, $\sim\!{2}\;{\rm ns}$ pulse duration, a smooth spectrum, and sufficient power in the 532–600 nm range. The supercontinuum laser light, after passing through a $\sim\!{15}\;{\rm m}$ free-space delay line, was focused on a rotating thin paper disc placed in the focus plane of the instrument objective. The paper acts as a near-isotropic scatterer with negligible temporal delay, while the rotation helps average out the interference fringes from the coherent laser light.
Due to the uncertainty in the relative timing between the calibration measurement and the actual measurements, a series of relative intensity calibration measurements was taken in steps of 50 ps additional delay (relative to the supercontinuum laser trigger signal). Additionally, using the standard setup, a similar series of measurements was performed in steps of 50 ps, to figure out the optimal time delay for rejecting short lifetime fluorescence, i.e., maximize the Raman signal SNR. For these measurements, multiple samples were used: coronene (strong, short lifetime fluorescence), cyclohexane (no fluorescence), methanol (no fluorescence), methanol with rhodamine (strong, short lifetime fluorescence), and a selection of natural mineral samples.
The optimal timings for the measurement and the relative intensity calibration were found by combining all this data in a two-step process. The first step is to find the measurement timing setting that maximizes Raman signal relative to the fluorescent background, as illustrated in Fig. 4 using the data from the coronene sample. The SNR is calculated based on the height of a strong Raman peak (e.g., at ${1353}\;{{\rm cm}^{-1}}$ for coronene) and the background (estimated next to the peak). As such, the calculated SNR number ignores any fixed pattern noise. A delay of 5900–5950 ps consistently maximizes the SNR for all samples with short lifetime fluorescence. For samples with long lifetime, or no fluorescence, such as cyclohexane, a delay of 6100 ps was optimal as it maximizes the collected Raman signal.
Fig. 4. (a) Short measurements of a coronene sample with different delay settings. Here, a relative intensity calibration optimized for a delay of 5950 ps has been applied to all spectra. Two Raman peaks are visible around ${1360}\;{{\rm cm}^{-1}}$, and an additional peak is visible at $\sim\!{500}\;{{\rm cm}^{-1}}$. The fixed pattern noise visible for later delay settings illustrates the need for an accurate relative intensity calibration that captures the pixel-to-pixel timing variations, in addition to the sensitivity and spectral response of the instrument. (b) The height of the Raman peak, marked in dashed line in subfigure (a), and the background, evaluated just left of the Raman peak. (c) The SNR of the main Raman peak as a function of the delay. The SNR is defined as ${\rm SNR}={\rm Raman}$ peak height/sqrt (background counts), i.e., ignores the fixed pattern noise.
The second step is to find the delay timing setting for the relative intensity calibration that minimizes the fixed-pattern noise for the measurement with the delay that optimizes SNR, as found in step one. This is illustrated in Fig. 4, for example, by looking at the large step seen in the $\gt\! {6000}\;{\rm ps}$ delay data at just over ${1000}\;{{\rm cm}^{-1}}$. At delay 5950 ps (purple line), the step is no longer visible, while at delays $\lt={5900}$, the step is present as a negative feature. Thus, the correct relative intensity calibration timing has been found for a measurement timing of 5950 ps. However, Fig. 4 also highlights two issues with this calibration method:
- 1. It is evident that the calibration is only optimal for a subrange of the full spectral range of the spectrometer. For example, the measurement with 5950 ps delay (purple) removes the fixed pattern noise in the $\lt\! {1700}\;{{\rm cm}^{-1}}$ range, while the 6000 ps delay measurement (green line) more minimizes the fixed pattern noise in the range above. This may be due to chromatic dispersion in the nonlinear fiber in the supercontinuum laser.
- 2. In the range where the fixed pattern noise is removed, a new fixed pattern emerges: a small increase in counts every eighth pixel. While not debilitating in the measurement illustrated here, it will eventually be a limiting factor when analyzing unknown samples with smaller Raman peaks. This behavior is linked to the timing circuit of the SPAD array, which works in blocks of 8 pixels. Still, it is unclear why the pattern is not removed by the relative intensity calibration itself. Ultimately, this is a detector limitation that would require a redesign of the SPAD array readout architecture.
Despite clear limitations, the presented calibration method is sufficient to remove the significant fixed-pattern noise resulting from the short time gate operation using the SPAD array. While the calibration method remains a limiting factor to the performance of the overall instrument, at this stage the performance is good enough to illustrate the advantage of the time-gated Raman technique with shorter laser pulses and time gating.
C. Sampling Strategy
The sampling strategy used in this work was chosen to emulate the particular challenges associated with operating a microscopic point-sampling instrument on another planetary body as discussed in Section 1.A. Specifically, this means that the samples are not prepared, and interrogation is performed without physical contact with the samples. The instrument has to be fully automated (no human operator in the loop), and, due to limited spacecraft resources (time, energy, bandwidth data storage capacity etc.), it is imperative that each measurement yield a usable spectrum. To this end, our instrument operates with constant laser settings (i.e., no variable attenuator to optimize the laser power for each sample, as is usually done in laboratory Raman instruments). Based on experimental studies, we have found that, for this laser spot size ($\sim\!{10}\;\unicode{x00B5}{\rm m}$ FWHM) and pulse duration (100 ps), 10 nJ is a safe pulse energy that will avoid ablative laser damage even in very dark and/or sensitive materials. As mentioned, the laser repetition rate was limited to 150 kHz, which leads to an average laser power of 1.5 mW, which is a safe power to avoid thermal damage to all relevant samples.
Each sample was placed on an XYZ-motorized stage near the focus of the instrument objective. This emulates a three-axis fine positioning system integrated into the instrument head in a more realistic implementation of an instrument for planetary exploration. For each sample, the two corners (in X and Y) bounding the sampling area were established manually. Within this bounding area, the software randomized a fixed number of sampling points. For each sampling point, the software moved the stages to bring the sampling point into the center of the objective, followed by an autofocus routine, based on maximizing the counts returned from the sample. Due to potential fluorescence bleaching, the autofocus routine repeated the focusing in increasingly smaller ranges to converge on the true focus. This was found more robust than relying on autofocus based on the integrated context camera, especially for dark samples where the contrast in the image is lacking (note: for a flight instrument, a routine that could optimize autofocus for dark samples, which are often important organic rich targets, should be explored. This was beyond the scope of this work). In case no focus was found, a new random location was chosen, and the process was repeated. Once focus was established, the measurement was initiated. Each measurement comprises ${65.5}\;\times\;{{10}^6}$ laser pulses, with a total acquisition time of $\sim\!\!{9}\;{\rm min}$ (note: this acquisition time would be reduced to under a minute by running the laser at a repetition rate of $\sim\!{1}\;{\rm MHz}$, assuming the sample can handle the thermal load). Following each measurement, a separate measurement of the dark counts was performed. This was done by collecting ${32.8}\;\times\;{{10}^6}$ frames with the delay set to enable the SPAD gate a few nanoseconds before the arrival of the laser pulse. This step is not strictly necessary to perform after each sample point, and can be done more seldom depending on the characteristics of the final system. All measurements were performed in a lab with the fluorescent overhead lights on, and no light shielding around the sample.
3. RESULTS
A. Improvements: Sample Damage and Fluorescence Rejection
Figure 5 illustrates one advantage of the HMC lasers compared with the conventional $Q$-switched DPSS lasers. When probing a dark sample, in this case silicon, high power density leads to ablation damage to the sample. This ablation damage sets in when the pulse energy exceeds a certain threshold. Testing a wide range of samples, we found that 10 nJ is a safe setting for the laser in our instrument. However, since the collected Raman signal is proportional to the average laser power, the low repetition rate of conventional DPSS lasers becomes a significant bottleneck. HMC lasers overcome this bottleneck by allowing much higher repetition rates, up to MHz, with constant pulse energy.
Fig. 5. Measurement on a pure silicon wafer sample using the HMC laser and a conventional $Q$-switched DPSS laser (TEEM Photonics). The strong broadband signal seen in the conventional DPSS laser measurement is indicative of ablation damage resulting from the high pulse energy. Despite the shorter pulse duration and higher average power (100 ps versus 600 ps), the HMC laser stays below the ablation damage threshold, and the silicon Raman spectrum is easily distinguishable. The spot size for both measurements was $\sim\!{14}\;\unicode{x00B5}{\rm m}$ FWHM.
Fig. 6. SNR versus gate delay for a HMC laser and a conventional DPSS laser. This measurement was performed using a highly fluorescent coronene sample. The sample location was fixed between the measurement, ensuring the exact same sample spot was measured. Furthermore, the pulse energy was kept low (1 nJ) to make sure there was no sample degradation. The shorter pulse duration of the HMC laser enables improved rejection of the fluorescence background, leading to significantly improved SNR.
The second major advantage of the HMC lasers, compared with the conventional lasers, is the improved rejection of fluorescence as illustrated in Fig. 6, which compares the SNR versus gate delay, under identical conditions for the HMC and the conventional DPSS laser. The HMC laser offers $\sim\!{2.5}\times$ improved SNR, due to the shorter pulse duration. Furthermore, the acquisition time for the HMC laser was significantly shorter due to the higher pulse repetition rate (150 kHz versus 40 kHz), as the measurements were set to collect a fixed number of pulses.
B. Measurements on Mars Analog Samples
A set of samples was prepared and thoroughly characterized for the purpose of assessing the performance of the time-resolved Raman instrument and its potential use for planetary mineralogy. These six samples represent four primary aqueous lithologies that are regarded as high priority exploration targets for Mars astrobiology. They represent environments that, on Earth, are favorable for microbial biosignature capture and preservation: limestone, chert, sulfate evaporite, and shale [15]. All of the time-resolved Raman measurements in this work were performed on bulk samples with unprepared surfaces in their naturally occurring form with the exception of samples 2 and 4, which were rough cut (but not further prepared) from larger samples. The geological context and petrography of these samples were extensively characterized via the following techniques performed on prepared petrographic thin sections of the same samples: optical photomicrographs, fluorescence imaging, estimates of total organic content based on imaging the grains and using imaging processing algorithms to determine kerogen abundance, and x-ray diffraction analysis of bulk powdered samples to determine mineralogy.
We refer again now to the earlier discussion on sampling strategy required for autonomous operation in a remote setting versus in the laboratory with an active user operator. In the laboratory setting, fully characterizing these samples required a knowledgeable human involved in everything from sample preparation to microscopy and analytical techniques. Petrographic thin sections were an essential part of determining the existence and nature of biosignatures captured in these samples. This work is designed to test the utility of a “blind” grid sampling technique on these samples so that we can accurately build expectations for the performance of Raman spectroscopy operating in this atypical mode where the user is not in the loop. We present our initial results here; however, a few points must be noted. (1) Because of the remaining calibration issues discussed earlier, we are still limited by instrument noise. Future improvements in time-resolved detectors and fast pulsed lasers are expected to improve signal-to-noise further. (2) Due to issues with reliability of the lasers, we were limited in the number of spectra that we were able to measure. So in practice, more spectra could be added to the grid to increase the likelihood of detection of more minor constituents. Nonetheless, the work described in this section can give us a good understanding of what we can expect to see when operating a Raman spectrometer remotely. It is also important to note that a mission using Raman spectroscopy is also likely to have other instruments onboard that can provide complementary information that can help piece together a better picture of the sample as a whole in the context of the mission. It is our intention here to clarify the information that can be contributed by the Raman instrument alone.
Six representative analog samples were characterized using the grid sampling method. Minerals were identified using the Raman classification method described in Cochrane et al. [31]. The results are summarized in Table 1. A subset of Raman spectra are shown in Fig. 7, where each unique spectral type that was observed is represented.
Table 1. Summary of Sample Details and Measurement Results after Autonomous Grid Sampling of All Six Samplesa
Fig. 7. Raman spectra from all six samples. For each sample, multiple-point spectra were taken as indicated in Table 1. For each sample, the spectra were then grouped into sets of similar spectra. This figure shows one of each spectral type that was observed, and it can be assumed that other spectra in the same set were indistinguishable from the representative spectra shown here.
An important point to note is that we get a usable Raman spectrum every time, in other words we do not lose any spectra to fluorescence interference or sample damage. This is an important achievement, and was made possible only by the inclusion of the HMC lasers in the instrument. In comparison, CW Raman was attempted on thin sections from this sample set with mixed results. Although quantitative results were not reported (e.g., how many spots had fluorescence issues out of the total measurements attempted), qualitative results reported high fluorescence significantly impeding measurements in samples 2 and 6; some fluorescence in samples 4, 5, and 7; and minor fluorescence in sample 1.
Raman is a useful technique for identifying fossilized organic matter in various stages of decomposition (kerogen). Kerogen is identified by two spectral features at ${1600}\;{{\rm cm}^{-1}}$ and ${1350}\;{{\rm cm}^{-1}}$. One or both of these features could be identified in all samples except for 2 and 4. These samples have low kerogen content (0.27% and 1%). Although these samples do not show kerogen spectral features, it is interesting to note that sample 5 (0.5% kerogen) does show one of the features even though it is lower in content than sample 4. This could be due to differences in the maturity of the kerogen, which affects the Raman intensity. It is important to note that, for samples with low kerogen content (that is, all but sample 1 with 24% kerogen), we are operating very close to the signal-to-noise limit of our system. This highlights the challenge of detecting kerogen in small quantities in natural rock samples. All six samples show signs of fossilized organic matter via detailed microscopic examination of thin sections. For example, four types of fossils were identified in sample 6 (plant matter, carbonate shells, foraminifera, and grains of immature organics surrounded by halos of mature kerogen). It is important to recognize that identification of the two kerogen peaks by Raman spectroscopy is not in itself sufficient for identifying organics of biological origin. Abiotic carbon sources have been shown to be indistinguishable from biotic carbon sources in Raman spectroscopy, for example see work by Pasteris et al. [32]. This is why it is crucial that other contextual information be collected to complement Raman spectroscopy. It is not clear that such fossilized matter, only identifiable after sample preparation and microscopic analysis, would be positively identified in a remote planetary setting. That said, the detection of kerogen peaks in Raman spectroscopy along with a known geological setting could be a good indicator that these samples warrant further study, perhaps even justifying sample return to a terrestrial lab. Another interesting result is that when we do see kerogen in our Raman spectra, we see it in between 93% and 100% of all point spectra. This indicates that it is fairly evenly distributed and is fine grained, which is in line with what has been reported on these samples through study. We suggest that this is to be expected for samples formed in depositional environments where biologic matter is likely to be preserved, which underscores the utility of this sample set as an analog set for the search for life on other rocky planets, especially Mars.
4. DISCUSSION
A. Applicability of Time-Resolved Raman in Planetary Science
We have shown here that by using time-resolved Raman with an appropriate laser pulse width and repetition rate, it is possible to reliably obtain Raman spectra 100% of the time on samples that would otherwise present challenges due to fluorescence interference. Our ability to use this technique for a planetary science instrument in the near term is unfortunately hindered by technology limitations on both the laser and detector. The performance of the HMC lasers is well suited for the application, and the laser package is small and robust enough for the demanding requirements associated with planetary exploration instruments. However, the reliability issue would have to be addressed. Similarly, the SPAD array exhibits many of the characteristics desirable for a low mass/volume/power instrument, as well as desirable performance for the specific application (gating fall time). However, it is clear from the presented results that we have pushed the performance of the detector to its limit, as we run into issues with the readout architecture (limiting the repetition rate of the laser to $\lt\! {500}\;{\rm kHz}$) and the associated calibration method (fixed-pattern noise limiting the SNR). It is expected that the readout architecture issue could be resolved in a next generation device. It is possible that the improvements in such a device would enable a simplified temporal calibration routine. A complicated calibration procedure is a highly undesirable feature for an instrument intended for planetary exploration, and the one presented here is not amenable to such an application.
Instead, as shown by, e.g., Hanke et al. [33], a more promising application of time-resolved Raman spectroscopy would be as a laboratory instrument, where constraints on mass, volume power, acquisition time, etc., are significantly alleviated. The benefits are significant—fluorescence rejection and the ability to operate in daylight without light shielding.
B. How Well Does Automated Raman Perform for Planetary Science Applications?
The utility of blind automated grid-sampled Raman spectra is of prime interest here, as it best represents what would take place on a remote planetary science mission. As discussed, fluorescence interference is cited as one of the main weaknesses in using Raman spectroscopy. We demonstrate here that by using time-resolved Raman with short laser pulses we can minimize this issue.
In order to appreciate the contributions of Raman spectroscopy to analysis of an unknown sample in a planetary environment, it is useful to look at how much information could be ascertained from Raman spectroscopy of our Mars analog sample set. From there it would be most useful to determine whether we are able to obtain enough useful information to advocate for this technique.
The first point to note is that the major aqueous minerals (quartz, calcite, and gypsum) are easily and consistently identified. This is in large part due to the fluorescence rejection offered by the time-resolved technique. The second important point to note is that distinctive kerogen signatures were easily identified in samples having a $\sim\!{1}\%$ or better kerogen content. This is also made possible by the fluorescence rejection capabilities of this technique. This is also good news in that kerogen represents an important marker of samples that may contain material of biotic origin. However, caution must be used not to overinterpret this result because abiotic carbon is indistinguishable by Raman. The case for biogenicity is most easily made when Raman signatures co-occur with microscale morphological evidence. Context is everything, and the question becomes how much context would be gained by microscopic imaging where no thin sections are prepared? In this sample set, microfossils were all identified in thin section and not in bulk. The third point is that we identify a few minor minerals but not nearly the wealth of information that is obtained by detailed human-directed analysis of prepared samples. This is likely due to a number of factors including the weak intrinsic Raman signatures of many of these minerals (e.g., pyrite) as well as the fine-grained nature of sedimentary rock samples. The fact remains that, in this work, grid sampling did not find all of the minerals that we know are present in these samples in small quantities, and many of these are in the category of poor Raman scatterers. It is possible that longer acquisition times and the collection of more points would have found these minerals; however, there would be a cost to mission resources, specifically time and power. We were in large part prevented from performing more experiments of this nature due to the reliability issues we encountered with the laser. It would be interesting to see how many points and what collection time would be required in order to get a rich enough set of data to interpret these samples as containing material of biological origin. It is possible that it would simply not be realistic.
Finally, in this work, we are still limited by artifacts resulting from the laser and detector that could not be calibrated out. The main limitation of the laser is the poor reliability and the relatively low repetition rate. The primary limitations in the detector are the limited fall time, which limits the fluorescence rejection efficacy, and the nonoptimal readout circuit, which results in residual fixed pattern noise after the temporal calibration has been applied. It is expected that both of these limitations could be addressed, now or in the near future, by redesigning the readout circuitry and manufacturing new devices with SPAD-optimized CMOS technology. The artifacts present in our spectra may have contributed enough noise to hide minerals with very weak Raman signatures and therefore small features that would be buried in the noise. While this is a possibility, operating near the edge of the signal-to-noise capabilities of this technique is problematic to consider, when one is counting on identifying minor mineral phases.
In evaluating the benefit of Raman point mapping, it is useful to compare it with other techniques such as mapping by infrared reflectance spectroscopy. Although Raman has the ability to provide more specific and unique mineral identifications in many cases, point mapping can be arduous when compared to infrared reflectance spectroscopy, which can be much faster. While infrared mapping most often analyzes a much larger area of the surface, methods exist for obtaining quantitative mineral abundances via deconvolution techniques coupled with mature databases of analog spectra. In addition to not suffering from fluorescence issues, it also has the advantage of being able to identify amorphous or poorly crystalline materials in some cases where Raman cannot [34,35]. One must decide based on the specific goals of the planetary mission whether incurring the time to map single microscopic points is worthwhile, and how many points would be sufficient to gain the required information to characterize the surface. It is likely that there is a place for each of these techniques in the future of planetary exploration.
Funding
National Aeronautics and Space Administration ((NRA) NNH14ZDA001N-PICASSO); Jet Propulsion Laboratory (Strategic University Research Partnership (SURP)).
Acknowledgment
SPAD development was performed at Delft University of Technology by the group of Professor Edoardo Charbon. Dr. Svetlana Shkolyar (Carnegie Institute, Washington) provided valuable assistance with sample preparation and petrographic analyses related to the terrestrial analog studies. Professor George Rossman (Division of Geological and Planetary Sciences, California Institute of Technology) provided valuable input and edits to this paper. Andreas Richter (Batop) provided assistance in understanding and optimizing the performance of the microchip lasers from Batop.
REFERENCES
1. L. A. Nafie, “Recent advances in linear and nonlinear Raman spectroscopy. Part XII,” J. Raman Spectrosc. 49, 1874–1906 (2018). [CrossRef]
2. K. Buckley and A. G. Ryder, “Applications of Raman spectroscopy in biopharmaceutical manufacturing: a short review,” Appl. Spectrosc. 71, 1085–1116 (2017). [CrossRef]
3. D. Bersani and P. P. Lottici, “Raman spectroscopy of minerals and mineral pigments in archaeometry,” J. Raman Spectrosc. 47, 499–530 (2016). [CrossRef]
4. F. Rull, S. Maurice, I. Hutchinson, A. Moral, C. Perez, C. Diaz, and The RLS Team, “The Raman laser spectrometer for the ExoMars rover mission to Mars,” Astrobiology 17, 627–654 (2017). [CrossRef]
5. L. W. Beegle, R. Bhartia, B. Carrier, L. DeFlores, W. Abbey, S. Asher, A. Burton, M. Fries, P. Conrad, S. Clegg, K. S. Edgett, B. Ehlmann, W. Hug, R. Reid, L. Kah, K. Nealson, T. Nelson, M. Minitti, J. Popp, F. Langenhorst, P. Sobron, A. Steele, N. Tarcea, R. Wiens, K. Williford, and R. A. Yingst, “The SHERLOC investigation for MARS 2020,” in 48th Lunar and Planetary Science Conference, LPI Contribution No. 1964, id.2839 (2017).
6. R. C. Wiens, S. Maurice, and F. R. Perez, “The SuperCam remote sensing instrument suite for the Mars 2020 Rover: a preview,” Spectroscopy 32, 50–55 (2017).
7. J. Blacksberg, E. Alerstam, Y. Maruyama, C. Cochrane, and G. R. Rossman, “A miniaturized time-resolved Raman spectrometer for planetary science based on a fast single photon avalanche diode (SPAD) detector array,” Appl. Opt. 55, 739–748 (2016). [CrossRef]
8. L. A. Haskin, A. Wang, K. M. Rockow, B. L. Jolliff, R. L. Korotev, and K. M. Viskupic, “Raman spectroscopy for mineral identification and quantification for in situ planetary surface analysis: a point count method,” J. Geophys. Res. Planets 102, 19293–19306 (1997). [CrossRef]
9. I. M. Chou and A. L. Wang, “Application of laser Raman micro-analyses to Earth and planetary materials,” J. Asian Earth Sci. 145(B), 309–333 (2017). [CrossRef]
10. S. Shkolyar and J. D. Farmer, “Biosignature preservation potential in Playa evaporites: impacts of diagenesis and implications for Mars exploration,” Astrobiology 18, 431–453 (2018). [CrossRef]
11. A. I. Alajtal, H. G. M. Edwards, M. A. Elbagerma, and I. J. Scowen, “The effect of laser wavelength on the Raman spectra of phenanthrene, chrysene, and tetracene: implications for extraterrestrial detection of polyaromatic hydrocarbons,” Spectrochim. Acta A 76, 1–5 (2010). [CrossRef]
12. S. W. Squyres, J. P. Grotzinger, R. E. Arvidson, J. F. Bell III, W. Calvin, P. R. Christensen, B. C. Clark, J. A. Crisp, W. H. Farrand, K. E. Herkenhoff, J. R. Johnson, G. Klingelhöfer, A. H. Knoll, S. M. McLennan, H. Y. McSween Jr., R. V. Morris, J. W. Rice Jr., R. Rieder, and L. A. Soderblom, “In situ evidence for an ancient aqueous environment at Meridiani Planum, Mars,” Science 306, 1709–1714 (2004). [CrossRef]
13. S. L. Murchie, J. F. Mustard, B. L. Ehlmann, R. E. Milliken, J. L. Bishop, N. K. McKeown, E. Z. Noe Dobrea, F. P. Seelos, D. L. Buczkowski, S. M. Wiseman, R. E. Arvidson, J. J. Wray, G. Swayze, R. N. Clark, D. J. Des Marais, A. S. McEwen, and J.-P. Bibring, “A synthesis of Martian aqueous mineralogy after 1 Mars year of observations from the Mars Reconnaissance Orbiter,” J. Geophys. Res. 114, E00D06 (2009).
14. B. L. Ehlmann, J. F. Mustard, S. L. Murchie, J.-P. Bibring, A. Meunier, A. A. Fraeman, and Y. Langevin, “Subsurface water and clay mineral formation during the early history of Mars,” Nature 479, 53–60 (2011). [CrossRef]
15. D. T. Vaniman, D. L. Bish, D. W. Ming, T. F. Bristow, R. V. Morris, D. F. Blake, S. J. Chipera, S. M. Morrison, A. H. Treiman, E. B. Rampe, M. Rice, C. N. Achilles, J. P. Grotzinger, S. M. McLennan, J. Williams, J. F. Bell III, H. E. Newsom, R. T. Downs, S. Maurice, P. Sarrazin, A. S. Yen, J. M. Morookian, J. D. Farmer, K. Stack, R. E. Milliken, B. L. Ehlmann, D. Y. Sumner, G. Berger, J. A. Crisp, J. A. Hurowitz, R. Anderson, D. J. Des Marais, E. M. Stolper, K. S. Edgett, S. Gupta, N. Spanovich, and MSL Science Team, “Mineralogy of a mudstone at Yellowknife Bay, Gale Crater, Mars,” Science 343, 1243480 (2014). [CrossRef]
16. J. D. Farmer and D. J. Des Marais, “Exploring for a record of ancient Martian life,” J. Geophys. Res. 104, 26977–26995 (1999). [CrossRef]
17. Mars 2020 SDT (2013), Committee members: J. F. Mustard, M. Adler, A. Allwood, D. S. Bass, D. W. Beaty, J. F. Bell III, W. B. Brinckerhoff, M. Carr, D. J. Des Marais, B. Drake, K. S. Edgett, J. Eigenbrode, L. T. Elkins-Tanton, J. A. Grant, S. M. Milkovich, D. Ming, C. Moore, S. Murchie, T. C. Onstott, S. W. Ruff, M. A. Sephton, A. Steele, A. Treiman, Report of the Mars 2020 Science Definition Team, 154 pp., posted July, 2013, by the Mars Exploration Program Analysis Group (MEPAG) (2013).
18. J. L. Bishop and E. Murad, “Characterization of minerals and biogeochemical markers on Mars: a Raman and IR spectroscopic study of montmorillonite,” J. Raman Spectrosc. 35, 480–486 (2004). [CrossRef]
19. D. D. Wynn-Williams and H. G. M. Edwards, “Proximal analysis of regolith habitats and protective biomolecules in situ by laser Raman spectroscopy: an overview of terrestrial Antarctic habitats and Mars analogs,” Icarus 144, 486–503 (2000). [CrossRef]
20. H. G. M. Edwards, I. B. Hutchinson, R. Ingley, J. Parnell, P. Vitek, and J. Jehlicka, “Raman spectroscopic analysis of geological and biogeological specimens of relevance to the ExoMars mission,” Astrobiology 13, 543–549 (2013). [CrossRef]
21. T. Frosch, N. Tarcea, M. Schmitt, H. Thiele, F. Langenhorst, and J. Popp, “UV Raman imaging—a promising tool for astrobiology: comparative Raman studies with different excitation wavelengths on SNC Martian meteorites,” Anal. Chem. 79, 1101–1108 (2007). [CrossRef]
22. E. A. Cloutis, D. Hewson, D. M. Applin, J. Mann, and S. A. Mertzman, “Raman and reflectance spectroscopy of serpentinites and related hydrated silicates: effects of physical properties and observational parameters, and implications for detection and characterization on Mars,” Planet. Space Sci. 159, 66–83 (2018). [CrossRef]
23. M. Gaft, Modern Luminescence Spectroscopy of Minerals and Materials (Springer-Verlag, 2005).
24. E. Eshelman, M. G. Daly, G. Slater, and E. Cloutis, “Time-resolved detection of aromatic compounds on planetary surfaces by ultraviolet laser induced fluorescence and Raman spectroscopy,” Planet. Space Sci. 119, 200–207 (2015). [CrossRef]
25. A. Steinmetz, D. Nodop, J. Limpert, R. Hohmuth, W. Richter, and A. Tünnermann, “2 MHz repetition rate, 200 ps pulse duration from a monolithic, passively Q-switched microchip laser,” Appl. Phys. B 97, 317–320 (2009). [CrossRef]
26. E. Mehner, A. Steinmann, R. Hegenbarth, H. Giessen, and B. Braun, “Stable MHz-repetition-rate passively Q-switched microchip laser frequency doubled by MgO:PPLN,” Appl. Phys. B 112, 231–239 (2013). [CrossRef]
27. B. Bernard, E. Mehner, D. Kopf, H. Giessen, and B. Braun, “26 ps pulses from a passively Q-switched microchip laser,” Proc. SPIE 8960, 89601D (2014). [CrossRef]
28. E. Mehner, B. Bernard, H. Giessen, D. Kopf, and B. Braun, “Sub-20-ps pulses from a passively Q-switched microchip laser at 1 MHz repetition rate,” Opt. Lett. 39, 2940–2943 (2014). [CrossRef]
29. The ChemCam Team, “The ChemCam instrument suite on the Mars Science Laboratory (MSL) rover: body unit and combined system tests,” Space Sci. Rev. 170, 167–227 (2012). [CrossRef]
30. Y. Maruyama, J. Blacksberg, and E. Charbon, “A 1024 × 8, 0.7 ns time-gated SPAD line sensor for laser Raman spectroscopy and LIBS in space and rover based planetary exploration,” IEEE J. Solid-State Circuits 49, 179–189 (2014). [CrossRef]
31. C. J. Cochrane and J. Blacksberg, “A fast classification scheme in Raman spectroscopy for the identification mineralogical mixtures using a large database with correlated predictors,” IEEE Trans. Geosci. Remote Sens. 53, 4259–4274 (2015). [CrossRef]
32. J. D. Pasteris and B. Wopenka, “Necessary, but not sufficient: Raman identification of disordered carbon as a signature of ancient life,” Astrobiology 3, 727–738 (2003). [CrossRef]
33. F. Hanke, B. J. A. Mooij, F. Ariese, and U. Böttger, “The evaluation of time-resolved Raman spectroscopy for the suppression of background fluorescence from space-relevant samples,” J. Raman Spectrosc. 50, 969–982 (2019). [CrossRef]
34. G. A. Swayze, R. N. Clark, A. F. H. Goetz, K. E. Livo, G. N. Breit, F. A. Kruse, S. J. Sutley, L. W. Snee, H. A. Lowers, J. L. Post, R. E. Stoffregen, and R. P. Ashley, “Mapping advanced argillic alteration at Cuprite, Nevada, using imaging spectroscopy,” Econ. Geol. 109, 1179–1221 (2014). [CrossRef]
35. R. N. Greenberger, J. F. Mustard, B. L. Ehlmann, D. L. Blaney, E. A. CLoutis, J. H. Wilson, R. O. Green, and A. A. Fraeman, “Imaging spectroscopy of geological samples and outcrops: novel insights from microns to meters,” GSA Today 25(12), 4–10 (2015). [CrossRef]
