Ben Rogersa, Jay Grateb, Brett Pearsona, Neal Gallagherc, Barry Wisec, Ralph Whittena, Jesse Adamsa*
a Nevada Nanotech Systems, 1315 Greg Street, Suite 103, Sparks, NV, USA 89431;
b Pacific Northwest National Laboratory, 902 Battelle Blvd., P.O. Box 999, MSIN K4-13, Richland, WA, USA 99352;
c Eigenvector Research, Inc., 3905 West Eaglerock Drive, Wenatchee, WA USA 98801
Nevada Nanotech Systems, Inc. (Nevada Nano) has developed a multi-sensor solution to Chemical, Biological, Radiological, Nuclear and Explosives (CBRNE) detection that combines the Molecular Property SpectrometerTM (MPSTM)—a micro-electro-mechanical chip-based technology capable of measuring a variety of thermodynamic and electrostatic molecular properties of sampled vapors and particles—and a compact, high-resolution, solid-state gamma spectrometer module for identifying radioactive materials, including isotopes used in dirty bombs and nuclear weapons. By conducting multiple measurements, the system can provide a more complete characterization of an unknown sample, leading to a more accurate identification. Positive identifications of threats are communicated using an integrated wireless module. Currently, system development is focused on detection of commercial, military and improvised explosives, radioactive materials, and chemical threats. The system can be configured for a variety of CBRNE applications, including handheld wands and swab-type threat detectors requiring short sample times, and ultra-high sensitivity detectors in which longer sampling times are used. Here we provide an overview of the system design and operation and present results from preliminary testing.
Keywords: molecular, property, spectrometer, explosives, radiation, sensor, detection, classification
1.1 A combination molecular property and gamma spectrometer system
In a comprehensive study of the risks and liabilities of maritime terrorism, Greenberg et al.1 concluded that “a dirty- bomb attack perpetrated using an illicit cargo container presents the greatest combination of likelihood and expected economic harm” to the United States. Based on recommendations made by the 9/11 Commission, Congress passed a law in 2007 specifying that no cargo container may enter the United States before being scanned with imaging equipment and a radiation-detection device; however, the U.S. has failed meet the July 2012 deadline for doing so, and extended the deadline by two years.2
Since 2004, DARPA, the DOD and the DOE have supported the early stage development of a sensor system for use in terrorist threat detection systems. This work has led to a prototype system combining a solid-state gamma spectrometer with the Molecular Property SpectrometerTM (MPSTM) technology. The sensor system can be configured for a variety of homeland security applications, and is currently designed to be individually mounted in shipping containers. This is the concept of operations envisioned by the Department of Homeland Security (DHS) for the Time Recorded Ubiquitous Sensor Technologies (TRUST) program—a compact, low-cost, low-power unit for detection of CBRNE threats, people, and contraband smuggled inside International Organization for Standardization (ISO) containers. The system, known as C-ScoutTM and shown in Figure 1, interfaces with the Marine Asset Tag Tracking System (MATTS) developed for DHS to securely transmit information to wireless hubs deployed in ports and highway checkpoints. The MATTS system can also communicate via cellular or satellite phone for remote data transmission.
System development has focused on detection of commercial, military and improvised explosives, radioactive materials, and chemical threats, with primary emphasis on explosives and radioactive materials. As such, at this stage in development the unit is best suited for detection of a radiological dispersal device (i.e., dirty bomb) that includes explosive material and a radioactive isotope.
2.1 Molecular Property SpectrometerTM (MPSTM) Subsystem
The Molecular Property SpectrometerTM (MPSTM) subsystem is implemented with a micro-electro-mechanical, chip- based technology capable of measuring a variety of thermodynamic and electrostatic molecular properties of sampled vapors and particles. Figure 2 shows the layout and micrographs of an MPSTM chip as well as the packaged chip mounted on a board. Various MPSTM chip designs have been built. The chips contain up to 18 sensing elements and can perform up to nine different types of measurements, essentially forming a “system of systems.”
The main sensing structure is based on a patented array of low-thermal-mass microcantilevers (nanobalances) with integrated piezoelectric sensing elements that provide actuation and sensing of resonance frequency.3 This allows an array of sensors to be electrically monitored in a low-cost, low-power, robust fashion rather than using an optical readout (which is more common but more expensive and less robust). The microcantilevers also have built-in resistive heaters for thermal analysis and also for cleaning each sensor after processing a sample. These built-in resistors also serve as temperature sensors to enable temperature and flow compensation in order to minimize noise and drift and further enhance sensitivity. In addition, the piezoelectric layer serves as an impedance sensor, used to provide electrical characterization of the sample or of the environment (such as humidity level). Up to seven of the sensors of the array are coated with semi-selective polymers, designed to classify molecules according to their solubility interaction with these polymers.4 In addition, up to three sensors can be coated with metals and one with silicon dioxide to provide additional chemical affinity surfaces.
The MPSTM subsystem analyzes molecular properties of absorbed and adsorbed molecules under proprietary test protocols and conditions. Specific techniques utilized include gravimetric (mass) analysis and a variety of thermal analysis techniques including Differential Thermal Analysis (DTA), Thermogravimetry (TG), and Differential Scanning Calorimetry (DSC). The core molecular properties underlying the MPSTM analysis include the full range of polymer- solubility interactions (dispersion interaction, polarizable, dipolar, H-bond basic and H-bond acidic), as well as measurement of diffusion rates, absorption and desorption rates, heat of vaporization, melting point, resistivity, capacitance, thermal conductivity, specific heat, activation energy, heat of reaction, heat of hydration, heat of dissociation, and boiling point. The wide array of complementary measurement techniques enables MPSTM technology to achieve higher specificity and lower false-positive rates than these techniques achieve when used separately.
The chip is mounted in a flow-through package that plugs onto the sensor electronics board, as shown in the lower right picture in Figure 2. To further maximize sensitivity, samples flow directly through a hole in the middle of the chip to maximize collection of sample on the sensors.
While microcantilever sensitivities have been demonstrated into the attogram mass range, we have achieved 20- picogram resolution with the current generation device, which is sufficient when combined with the preconcentrator to make parts-per-trillion level measurements of explosives. We are targeting 250-femtogram resolution in future systems.
The detection process occurs in four stages: 1) Sample Collection, 2) Pre-concentration, 3) Pre-separation, and 4) Sample analysis. These are each discussed in further detail below.
Stage 1—Sample Collection: The MPSTM detection and analysis process begins by collecting a sample. Like currently available trace detection systems, this sample is typically a sample of the air—be it from the inside a shipping container or from the vicinity of an object of interest such as a person, package, or a swab. Sample collection is performed with the aid of a pump which pulls the sample into the Concentration-Separation (Con-Sep) module. In the event that a particle is collected, Stage 3 of the analysis provides heat to create vapors from the particle for analysis.
Stage 2—Analyte Concentration: Samples first pass into the Con-Sep module, which is a preconcentrator tube packed with absorbent material (e.g., Tenax TA, or a fine metal mesh) that traps vapors and particles. Tenax TA has a high capture efficiency for a broad spectrum of vapors, including explosives,5 a low affinity for water, and a long useful life (hundreds to thousands of cycles). Metal mesh has an even longer life as well as a high capture efficiency for many explosives. Preconcentration factors vary by chemical and depend on sampling time and sampling rate, but are typically about 100X for short collection times. Preconcentration factors up to ~4000X have been achieved for nerve agent simulant DMMP and explosives like TNT and PETN. The Con-Sep preconcentrator allows the system to collect macroscopic sample volumes (liters and above) and condenses them down to milliliter-scale volumes for analysis by the sensor chip. This can improve the sensitivity of the system by three or more orders of magnitude depending on the time constraints of the measurement.
Stage 3—Sample Separation: Trapped molecules are sequentially released by increasing the temperature of the Con-Sep in a controlled fashion. Physical parameters including vapor pressure and boiling point affect the temperature (time) at which each vapor will elute and be carried downstream to the sensor chip. Such separation enables the molecular properties of a complex sample to be measured in a sequential fashion, thereby reducing the complexity of the analysis. This operation is analogous to the function performed by a gas chromatograph column in a Gas Chromatography-Mass Spectroscopy (GC-MS) system. The Con-Sep preconcentrator technology was initially developed, tested, and proven at Pacific Northwest National Labs6 and has been licensed by Nevada Nano.
Stage 4—On-Chip Analysis of Molecular Properties: The molecular property information is collected continuously for each vapor sample to provide analysis of the (separated) stream of molecules emitted from the Con-Sep. External electronics collect this data. Thousands of individual measurements are made on a typical sample, creating a rich data set to identify the sample components. This is done using custom chemometric, pattern recognition software to compare the measured results to a stored chemical database. The MPSTM system can be taught to identify new molecules by adding the data from those substances to the database.
In theory, the measurement of fundamental properties of molecules allows the system to measure any chemical— including those emanating from micro-organisms and humans—thus providing the capability to address the broadest possible spectrum of threats. The capabilities of this approach to identify a variety of chemical, biological and explosive threats have been partly demonstrated in lab testing in a variety of complex backgrounds. The MPSTM has been used to analyze many substances, including several Toxic Industrial Chemicals, a nerve agent simulant, a bio-warfare agent simulant, bacteria, a variety of common substances (perfumes, cleaning solutions, tobacco, etc.) and the explosives TNT, RDX, TATP, HMTD, and PETN.
Highly orthogonal (i.e. diverse) measurement systems have the potential to greatly improve the performance of Trace Detection Systems7,8 by reducing the number of false positives (due to the additional data collected from orthogonal measurements) but possibly also reducing the number of false negatives (in the event that data from some channels is inconclusive and an identification can be made based on the strength of the data in other channels). These improvements would result from not relying on the measurement of just one physical property to identify a threat but instead using multiple property measurements.
While orthogonal measurement capability has been desired for some years, the complexity of making multimode measurement systems has impeded their development. By integrating multiple systems on a “Self-Sensing ArrayTM” type of chip (cantilevers with integrated transduction mechanisms), Nevada Nano has been able to overcome much of the multisystem complexity. Utilizing a microelectromechanical system (MEMS) process of very modest complexity, we have integrated temperature sensing, heat control, impedance measurement structures and mass measurement into a robust, solid state electronic tool that provides a system of systems on a chip. Still, not all of the complexities and trade- offs of making these measurements on a single chip are fully understood. One key challenge in this development involves taking advantage of all the possible measurements of the MPSTM, and utilizing chemometric techniques to interpret the large amount of data generated.
Results and discussion of chemometrics performed on the variety of data produced by this prototype system will be the subject of subsequent papers. Up to now, we have begun developing initial SIMCA (Soft Independent Modeling of Class Analogy) models for TG and DTA measurements, utilizing a subset of features extracted from initial data. SIMCA is based on Principal Components Analysis (PCA), the statistics and nomenclature of which are provided elsewhere.9,10 As data are generated, revised models can be quickly parameterized and implemented. Parameterization includes determining classification parameters (e.g., using Q, T2 or a combination, setting decision limits), setting decision rules based on multiple instrument responses (e.g., dTG and DTA), determining optimal data preprocessing (e.g., normalization) and selecting discriminating features for each measurement. The objective for parameterization is to provide accurate classification of targets without false alarming on interferent substances. We are also developing additional classification algorithms in parallel in an effort to identify which approach will provide robust and reliable results for long-term instrument operation. One approach will be based on the classical least squares (CLS) model for target detection and classification. In contrast to the SIMCA model that used features and statistics extracted from raw data, the CLS-based model will be used on the raw data itself. An anticipated advantage of the CLS model, which will be considered a target detection algorithm, is that we expect to be able to employ non-negativity constraints. Both SIMCA and CLS require data from a well-implemented design of experiments to allow parameterization.
2.2 Gamma spectrometer module
Detection of radioactive materials intended for use in a nuclear weapon or a dirty bomb depends on detection of high- energy gamma radiation and/or neutrons emitted from the hidden device. Multiple factors determine the amount of radiation that can be detected in a given scenario: the mass and type of material, the type and amount of shielding, the background radiation from natural and man-made sources, the sensor’s capture efficiency and size, the source-to-sensor distance, and the collection time.
The gamma spectrometer module in the C-ScoutTM (Figure 3) includes a solid-state detector, a low-power ASIC, and a microprocessor. It detects radiation in the range of 30 keV-3MeV with a resolution of 1.5% to 2.5% FWHM at 662 keV, which is superior to comparable units using other radiological detection materials: NaI (6-7%), BGO (9.5-11.7%), CsI (8-9%), CWO (7.3-9%), GSO (7.8-8.9%), LaBr3 (2.8-3.6%), LaCl3 (5%), LGSO (7.8%), LSO (8.8-10.3%), LYSO (8.9%), MLS (83.4%), PVT (180%), SrI2:Eu (2.6-5.3%), and Cs2LiYCl6:Ce (5.1%).11, 12, 13, 14, 15, 16, 17, 18 The solid-state detector is made from a 1-cm3 crystal of cadmium zinc telluride (CZT). The unit requires ~180 mW of power in full spectroscopy mode.
Spectroscopic data gathered by the solid-state detector contain peaks corresponding to photon energies from gamma rays or x-rays derived from radioactive isotopes in the detector’s vicinity or from secondary processes related to the initial nuclear decay of these isotopes. These data are analyzed by custom isotope identification software, which compares the spectrum to known spectrum libraries and outputs the types of isotopes detected and the relative confidence of that identification. The program is currently capable of recognizing 37 isotopes, including likely dirty bomb materials (e.g., 137Cs, 241Am) and nuclear weapon materials (e.g., 103Pd, 238Pu, 239Pu, 234U, 235U, 238U). The software also looks for higher-than-normal radiation levels—which can serve as a warning in the case that an isotope is not identified specifically—as well as x-rays indicative of lead-shielded materials. It includes temperature-compensation capability and is easily expandable to include additional isotopes.
The radiation module does not currently incorporate the direct neutron sensing preferred for detection of nuclear material/weapons (e.g., using a Helium-3-based sensor). Still, certain types of nuclear material can be detected via gamma radiation. Our model suggests the current gamma spectrometer could detect 15 pounds of Highly Enriched Uranium, 97%-shielded, located 40 feet away in the opposite corner of a shipping container within 10 minutes.* A plutonium-based bomb is significantly more difficult to detect using gamma radiation alone, owing to its smaller mass (roughly 4 kilograms are needed to construct a weapon)19 and its relatively low-energy, low-yield photons. While our model predicts the unit could detect such a weapon at a distance of 40 feet within seconds if unshielded, a thin (1-mm) sheet of lead would shield the device by 99.999999996% and make gamma-based detection all but impossible. However, the module can also detect secondary gamma emissions created by neutron interactions with various materials upon absorption of neutrons.20
The lead used to shield gamma radiation does not effectively shield neutrons. (Other neutron-shielding materials would be required, such as borated polyethylene, hydrogen, cadmium, boron, water, paraffin, or concrete.) Neutron detection is the best approach for certain nuclear materials such as plutonium. As such, the current module’s capability could be expanded beyond gamma-radiation-based detection (which is best suited for the detection of dirty bombs and the illicit trafficking of many radioactive materials) by integrating the best-available commercial off-the-shelf (COTS) neutron detector and corresponding electronics. This would likely be one of the developing technology alternatives to Helium-3- based detectors, which are being phased out due the shortage of He-3.21
Differential Thermal Analysis (DTA) has been used for decades as a means of examining the thermal properties of explosives.22 The MPSTM has the ability to utilize this and other thermal analysis techniques to analyze explosive materials at smaller size scales and faster time scales than those typical in conventional thermal analysis. Here we focus our discussion on lab and field results from the DTA measurement technique.
3.1 Differential Thermal Analysis (DTA)
Conventionally, DTA involves heating or cooling a test sample and an inert reference under identical conditions, while recording any temperature difference between the sample and reference. This differential temperature is then plotted against time, or temperature. We use a variation of this technique for conducting DTA with the MPSTM in which heat is rapidly applied to a cantilever and the temperature response is measured using the cantilever’s integrated thermistor. We then compare the temperature response of a cantilever exposed to a sample to that of a clean cantilever (the comparison creating a differential analysis). When a sample is present, the measured signal is caused by endothermic and exothermic events that can include melting, evaporation and decomposition—producing a distinct temperature profile for each analyte. Total analysis time is about one second. This technique has been shown to produce different signals, or signatures, depending on the type of molecule on the sensor.23 These unique shapes in the resistance response are then used as the identification method for this channel. Figure 4 shows representative DTA responses from three explosives.
3.2 Field Testing with the C-ScoutTM Prototype
A C-ScoutTM prototype was installed inside a standard, 40-foot shipping container in the Nevada Nano parking lot. The system was controlled by a laptop via a local network setup inside the container. Common cargo was simulated in the container with plastic-wrapped cardboard boxes on wood pallets. This created a background headspace vapor inside the container due to outgassing of these common shipping materials. During a typical test, the system sampled the container headspace at a rate of 350 ml/min for 1 to 2 hours, while simultaneously collecting gamma radiation spectra. A Con-Sep incorporating a 5-um metal mesh was used.
Data were stored on the C-ScoutTM unit and later transferred via Ethernet cable to a tethered laptop for analysis. The protocol used included a self-calibration step in which the sensors were interrogated twice—once with collected analyte and once after the system’s automated cleaning step—with the second of these two analyses serving as a reference to compensate for temperature and humidity effects, run to run. The container internal temperature ranged from -5 to 30C over the course of this testing campaign, and relative humidity ranged from 10 to 57%.
Ground truth testing performed by independent analysis (using EPA method TO-17) showed that the container headspace contained approximately two dozen compounds, for a combined “background” vapor concentration of ~42.5 ppb. See Table 1.
To safely simulate a dirty bomb, testing was performed using a small metal plate as a TNT vapor source and a radiation source puck containing 11 nanograms of the isotope Cesium-137 (137Cs). To create the TNT source, 100 mg of flake TNT was dissolved in methanol solvent and spread over a 12 x 19 cm metal plate. The plate was positioned face-up on a hotplate (to control its temperature) near the C-ScoutTM intake. Again, independent ground truth analysis was performed to determine the TNT vapor concentration collected at the C-ScoutTM intake using this configuration. The results showed the TNT concentration to be <340 parts per trillion. This was based on <75 ng TNT collected in 24 liters of container headspace, since 75 ng TNT was the limit of detection for the instrument and protocol used (OSHA Method 44). Figure 6 shows representative results of Differential Thermal Analysis (DTA) after a 60-minute sample collection. The low-volatility background vapors create the peak in the DTA signal near x=0.075 sec. When the <340-ppt TNT vapor is also present, a unique shoulder peak forms near x=0.067 sec on the background-based peak, as shown in Figure 6. The shape and location of such TNT peaks correlate with TNT peaks from other lab and field tests, as has been confirmed with initial chemometric analysis. [caption id="attachment_1883" align="alignnone" width="1048"] Figure 6. Representative DTA responses during field testing inside the shipping container to collections of container air only, and container air with the TNT source present. The container air “background” contained ~42.5 ppb from ~24 components (see Table 1). The responses shown here represent the summed DTA output of three cantilever sensors. Signals have been smoothed, filtered, leveled and zeroed at t=0.05 sec.[/caption]
Meanwhile, during these TNT tests, a 1-μCi (11-nanogram) Cesium-137 source was placed 62.5 cm away from the C- ScoutTM. This activity level and distance were used to simulate the sensitivity requirements of IEEE/ANSI standard N42.35 “Evaluation and Performance of Radiation Detection Portal Monitors for Use in Homeland Security.”
Figure 7 displays spectrum data from eight representative two-hour tests, four with the1-μCi 137Cs source present and four without. Detection and identification were performed automatically by the C-ScoutTM during analysis using custom- designed spectrographic pattern recognition software that runs in tandem with the explosives detection protocol. Note the characteristic 662-keV photon peak from 137Cs. The broad peak centered near 150 keV in the spectrum plots is known as the “Compton shelf,” an ever-present artifact of photon detectors due to Compton scattering. Potassium-40 (40K) was often also identified during field tests. This is expected, as 40K is the most common naturally occurring radioactive element on earth, with a well-known 1460.8- keV gamma ray found ubiquitously in background spectra.24
In applications where a small, low power, stand-alone CBRN&E system is required, we believe a multi-transducer array has the highest chance of success because multiple, orthogonal measurement techniques are brought to bear to enhance selectivity and reduce false positives. The micro-electro-mechanical, chip-based technology discussed in this paper was designed to prove out this hypothesis. Having combined the solid-state Molecular Property SpectrometerTM (MPSTM) technology with a gamma spectrometer in a single system and conducted preliminary testing of its performance, we believe this approach to trace detection of CBRN&E materials is a good choice for autonomous threat detection systems.
 Greenberg, M., Chalk, P., Willis, H., Khilko, I., and Ortiz, D., “Maritime Terrorism: Risk and Liability,” RAND Center for Terrorism Risk Management Policy (2006)
 Frantz, D., “Port security: U.S. fails to meet deadline for scanning of cargo containers,” Washington Post, 15 July 2012, http://articles.washingtonpost.com/2012-07-15/world/35489894_1_port-security-port-vulnerability-cargo- containers
 Minne, S., Manalis, S., and Quate, C., “Parallel atomic force microscopy using cantilevers with integrated piezoresistive sensors and integrated piezoelectric actuators,” Appl. Phys. Lett, vol. 67, no. 26, 3918-3920 (1995)
 Grate, J., “Acoustic Wave Microsensor Arrays for Vapor Sensing,” Chem. Rev. 100, p. 2627 (2000)
 Sigman , M., Ma, C.-Y., and Ilgner, R., “Performance Evaluation of an In-Injection Port Thermal Desorption/Gas Chromatographic/Negative Ion Chemical Ionization Mass Spectrometric Method for Trace Explosive Vapor Analysis,” Anal. Chem. 73, 792-798 (2001)
 Grate, J., Anheier, N., and Baldwin, D., “Progressive Thermal Desorption of Vapor Mixtures from a Preconcentrator with a Porous Metal Foam Internal Architecture and Variable Thermal Ramp Rates,” Anal. Chem. 77(6), 1867-1875 (2005)
 Weimar, U., and Göpel, W., “Chemical imaging: Trends in multiparameter sensor systems,” Conf. Proc. EUROSENSORS XI, Warsaw (1997)
 Bonne, U., Eden, G., Frye-Mason, G., Sacks, R., and Synovec, R., “Propulsion and Power Rapid Response R&D Support, Delivery Order 0013: Micro Gas Chromatography Tradeoff Study,” AFRL-PR-WP-TR-2004-2060 (November 2003)
 Wise, B.M. and Gallagher, N.B., “The Process Chemometrics Approach to Chemical Process Monitoring and Fault Detection,” J. Proc. Cont. 6(6), 329-348 (1996)
 Jackson, J.E., [A User’s Guide to Principal Components], John Wiley & Sons, New York, NY (1991)
 Nassalski, A., “Comparative Study of Scintillators for PET/CT Detectors,” IEEE Transactions on Nuclear Science, 54(1) 3-10 (2007)
 Avdeichikov, V., “Light output and energy resolution of CsI, YAG, GSO, BGO and LSO scintillators for light ions,” Nuclear instruments & methods in physics research, Section A, Accelerators, spectrometers, detectors and associated equipment, 349(1) 216-224 (1994).
 Shah,K., Glodo, J., Klugerman, M., Moses, W., Derenzo, S., and Weber, M., “LaBr3:Ce Scintillators for Gamma Ray Spectroscopy,” IEEE Transactions on Nuclear Science, 50, 2410-2413 (2003)
 Wan, W.-H., “The Operational Characteristics of a Sodium Iodide Scintillation Counting System as a Single- Channel Analyzer,” Radiation Protection Management, Volume 20, No. 5 (2003)
 Cherepy, N., Payne, S., Hawrami, R., Burger, A., Boatner, L., Van Loef E., and Shah, K., “Prospects for High Energy Resolution Gamma Ray Spectroscopy with Europium-Doped Strontium Iodide,” Materials Research Society Symposium Proceedings ISBN 978-160511137-7, Volume 1164, Pages 179-184 (2010)
Wilson, C., van Loef, E., Glodo, J., Cherepy. N., Hull, G., et al., “Strontium iodide scintillators for high energy resolution gamma ray spectroscopy”, Proc. SPIE 7079, Hard X-Ray, Gamma-Ray, and Neutron Detector Physics X, 707917 (September 05, 2008)
 Glodo, J. Brys, W. Entine, G. Higgins, W.M. van Loef, E.V.D. Squillante, M.R. Shah, K.S., “CS2LiYCl6: Ce Neutron gamma detection system,” Nuclear Science Symposium Conference Record, NSS ’07. IEEE, Vol 2, p. 959 (2007)
 Cochran, T., and Paine, C., “The Amount of Plutonium and Highly-Enriched Uranium Needed for Pure Fission Nuclear Weapons,” Natural Resources Defense Council, (13 April 1995), Web: http://www.nrdc.org/nuclear/fissionw/fissionweapons.pdf
 Mitchell, D., Harding, L., and Smith, K., “Neutron Detection With Gamma-Ray Spectrometers for Border Security Applications,” IEEE Transactions on Nuclear Science, Vol. 57, No. 4, p. 2215 (Aug. 2010)
 “Neutron detectors: alternatives to helium-3,” United States Government Accountability Office, Report to Congressional Requestors, (September 2011)
 Collins, L. and Haws, L., “The Thermochemistry of Explosives: A Review,” Thermochimica Acta, 21, p. 1-38 (1977)
 Senesac, L., Yi, D., Greve, A., Hales, J., Davis, Z., Nicholson, D., Boisen, A., and Thundat, T., “Micro-differential thermal analysis detection of adsorbed explosive molecules using microfabricated bridges,” Rev. Sci. Instrum., 80, 035102 (2009)
 Gaylord, R., “Radioactivity of Potassium Solutions: A Comparison of Calculated Activity to Measured Activity from Gross Beta Counting and Gamma Spectroscopy,” Lawrence Livermore National Laboratory, UCRL-TR-214061 (July 26, 2005)