Billion-pixel X-ray camera (BiPC-X)
Zhehui Wang, Kaitlin Anagnost, Cris W. Barnes, D. M. Dattelbaum, Eric R. Fossum, Eldred Lee, Jifeng Liu, J. J. Ma, W. Z. Meijer, Wanyi Nie, C. M. Sweeney, Audrey C. Therrien, Hsinhan Tsai, Xin Que
BBillion-pixel X-ray camera (BiPC-X) a) Zhehui Wang, Kaitlin Anagnost, Cris W. Barnes, D. M. Dattelbaum, Eric R. Fossum, Eldred Lee,
2, 1
JifengLiu, J. J. Ma, W. Z. Meijer, Wanyi Nie, C. M. Sweeney, Audrey C. Therrien, Hsinhan Tsai, and Xin Yue Los Alamos National Laboratory, Los Alamos, NM 87545, USA Dartmouth College, Hanover, NH 03755, USA Gigajot Technology, Pasadena, CA 91107, USA Universit´e de Sherbrooke, Sherbrooke, QC J1K 2R1, Canada (Dated: 7 January 2021)
The continuing improvement in quantum efficiency (above 90% for single visible photons), reduction in noise(below 1 electron per pixel), and shrinking in pixel pitch (less than 1 micron) motivate billion-pixel X-raycameras (BiPC-X) based on commercial CMOS imaging sensors. We describe BiPC-X designs and prototypeconstruction based on flexible tiling of commercial CMOS imaging sensors with millions of pixels. Devicemodels are given for direct detection of low energy X-rays ( <
10 keV) and indirect detection of higher energiesusing scintillators. Modified Birks’s law is proposed for light-yield nonproportionality in scintillators as afunction of X-ray energy. Single X-ray sensitivity and spatial resolution have been validated experimentallyusing laboratory X-ray source and the Argonne Advanced Photon Source. Possible applications include widefield-of-view (FOV) or large X-ray aperture measurements in high-temperature plasmas, the state-of-the-artsynchrotron, X-ray Free Electron Laser (XFEL), and pulsed power facilities.
I. INTRODUCTION
Room-temperature Complementary Metal OxideSemiconductor (CMOS) imaging sensors have enteredthe single-visible-photon-sensitive regime withoutavalanche gain . Uses in personal devices such as cellphones and growing applications in machine vision havecontinuously pushed performance improvements, Fig. 1,and cost reduction for CMOS imaging sensors (CIS). Asa result, CIS have gradually taken over charge coupleddevices (CCD) imaging sensors over the last decade.Compared with CCD, which are serial devices whenlight-induced charge is read out one pixel at a time,row/column by row/column, CIS are based on parallelpixel architecture, when all pixels are designed to beexactly the same, including the readout electronics.Since electric charge from each pixel can be read out inparallel, CIS are better suited for high-speed applicationsthan CCD. Consumer CIS have already reached 1000frames per second (fps). One of the main results hereis that high-performance low-cost visible-light CIS opendoor to billion-pixel X-ray camera (BiPC-X) designs,which may find applications such as in wide field-of-viewmeasurements of high-temperature plasmas, pulsedpower facilities, and X-ray scattering experiments in thestate-of-the-art light sources including synchrotrons andX-ray free electron lasers. There are several approachesto overcome the low detection efficiency of the visible-light CIS for X-ray photon detection. A multi-layerCIS architecture has been described recently , andvalidated with initial X-ray experiments at the Argonne a) Contributed paper to the Proceedings of the 23rd Topical Con-ference on High-Temperature Plasma Diagnostics, Santa Fe, NM,USA, May 31 - June 4, 2020. Rescheduled online, Dec. 14-17, 2020.Correspondence: (Z.W.) [email protected].
Advanced Photon Source (APS) . Another approach isto integrate photon energy attenuation layers (PALs)with CMOS at pixel level . Alternatively, we mayenhance the X-ray efficiency of each CIS by a scintillatorconverter. The latter approaches can also be extendedto a multilayer configuration. FIG. 1. A brief survey of the evolutionary trends of CCD (inblue) and CIS (in red) over the last 25 years. The quantum ef-ficiency (QE) for visible photons has now exceeded 90%. Thenoise level per pixel continues to decline, reaching 1 electronper pixel per readout cycle or less. Individual pixel size orpitch is < µ m as of 2020. These performance trends, incombination with continuing decline in cost, allow flexibilityin BiPC-X camera designs and applications. X-ray Bremsstrahlung and characteristic line emissionsfrom impurity ions are signatures of keV and higher tem-perature plasmas. Recent advances in data-driven sci-ence offer new toolboxes such as neural networks to diag-nose and understand high-temperature plasmas throughthree-dimensional (3D) X-ray imaging and tomography.Diffusive X-ray emissions from plasmas and the need to a r X i v : . [ phy s i c s . i n s - d e t ] J a n capture a large amount of X-rays data for applicationssuch as training of deep neural works motivate BiPC-X or a giga-pixel X-ray camera instrument. One of thefirst giga-pixel cameras, AWARE-2, was reported in 2012for visible light imaging . AWARE-2 used a 16-mm en-trance aperture to capture one-gigapixel images at threeframes per minute. The Large Synoptic Survey Telescope(LSST) camera has 3.2 billion pixels by tiling 189 CCDsand a 0.5-fps frame rate. A growing number of billion-pixel visible cameras has since been reported.Here we describe the design studies and initial resultstowards a BiPC-X. Sec. II is on the designs based ontiling of commercial CIS with millions of pixels and pro-totype construction using 3D printing of multi-sensorframe. In Sec. III, device models are given for directand indirect detection of X-ray photons. It is found thatabove 10% efficiency can in principle be obtained usingthe CMOS photo-diodes directly for photon energies be-low 10 keV. Modified Birks’s law is proposed for scintil-lator light yield. Sec. IV summarizes the experimentalresults on sensitivity and resolution. Follow-on work in-cludes application in plasmas and further optimization ofBiPC-X prototype design and performance. II. DESIGN & PROTOTYPE
Using as building blocks the CIS with millions of pixels(MP), a BiPC-X can be constructed through multi-layerstacking and tiling . Several possible configurations areillustrated as D , D and D in Fig. 2. The planar com-pact tiling configuration D increases the X-ray detec-tion aperture, which is proportional to the number ofCIS and the individual sensor area. The stacked tilingconfiguration D increases the aperture for high-energyX-rays above 20 keV that can penetrate through multi-ple layers of CIS. High-energy X-rays and gamma rays( ∼ MeV) are expected from run-away electrons in toka-maks and by nuclear fusion. Configuration D can beused in a toroidal plasma device such as a tokamak ora stellarator. The synchrotron radiation from run-awayelectrons in a torus, as well as the bulk X-ray emissionscan be captured by the CMOS sensor arrays surroundingthe plasma in the poloidal plane.There are a large number of commercial CIS to choosefrom, and they differ in the total number of pixels, pixelpitch, speed, and cost. The latest models offer 10s of MP.Examples include Samsung’s ISOCELL Bright HMX sen-sor (108 MP), the Canon 120MXS (122 MP), Gpixel’sGMAX3005 (150 MP), OmniVision’s OV64C (64 MP),and ON Semiconductor’s XGS 45000 (44.7 MP). A 5 × µ m except for GMAX3005(5.5 µ m, rolling shutter) and XGS 45000 (3.2 µ m, globalshutter). The frame rate of such a BiPC-X would be lim-ited to about 1k fps for now, depending on the CIS. Forpinhole imaging and tomography of inertial fusion plas-mas, the kfps frame rate of such a camera can be com- pensated by a.) Using the gated scintillator and micro-channel plate (MCP) frontend; or b.) Exposure timegating of the CIS. In both cases, one or several cameraswould capture one fast (1 µ s or shorter exposure time)X-ray image. A fast X-ray movie would be generated bygating the sensors with different pre-programmed timedelays. Additional customization of the CIS may be pos-sible by increasing the X-ray sensitive region. The directX-ray detection efficiency of the commercial off-the-shelfCIS is below 10%, limited by the pinned photodiode di-mension in each pixel to 2 - 5 µ m (the photodiode depthshould be larger than 3 µ m to ensure red sensitivity ) andthe small CMOS operating bias voltage of several volts .There are rooms to substantially increase the photodiodedepth to hundreds of microns, making such a photodiodeefficient for X-ray energies up to 10 keV, and thus sufficefor many laboratory high-temperature plasmas. CIS arecurrently manufactured on 200 mm to 300 mm Si wafers.A standard 200 mm silicon wafer has a thickness of 725 µ m. A 300 mm silicon wafer has a thickness of 775 µ m.Current visible light CIS only use a small fraction of thewafer thickness, less than 10 µ m. (a)(b) FIG. 2. (a) A BiPC-X may find applications in X-ray diffrac-tion (A), inertial confinement fusion (B) and magnetic fusion(C). Examples of stacking and tiling to form a BiPC-X: Pla-nar compact tiling configuration (D ), stacked tiling (D ), anddistributed tiling (D ). (b) A laboratory 2 × A laboratory 2 × µ m pitch, mono, die thickness 750 µ m, glasslid thickness 550 ± µ m) has been built, Fig. 2b. Weused a 3D printer (Lulzbot Taz 6) to make the mount-ing frame for the 4 CIS. The Fused Filament Fabricationprinting method used PolyMax PLA filament (from Poly-maker). The thickness of the frame printed is 0.100 (cid:48)(cid:48) toallow the detector to slightly protrude beyond the frame.Although the base circuit board is a 1.27 (cid:48)(cid:48) square, theimaging detector is slightly rectangular and offset fromthe center of the chip. This requires consideration of howdetectors will be oriented (for any size array) to ensurethere is adequate room for the attached circuitry. Asthey are now, the imaging detectors are required to beat least 0.32 (cid:48)(cid:48) apart to allow room for the boards theyare attached to without overlapping with one another.Using the Lulzbot Taz 6 printer, a monolithic frame forup to 8 × III. DEVICE MODELS
Here we describe device models for single X-ray photondetection efficiency and sensitivity. The response timeis currently limited at CIS. The analysis provides the-oretical basis for BiPC-X component selection and un-derstanding of the component testing data described inFig. 3 and in Sec. IV, especially the CIS and scintillators.Overall system performance parameters such as detectivequantum efficiency (DQE), resolution or blur character-ized by modulation transfer function (MTF) may alsobe derived, which is not included below partially due tothe observation that the X-ray source properties, X-raysource, object and detector standoff distances could alsoplay a role and thus need additional setup information .Device models may be divided into direct detectionschemes based on X-ray attenuation in silicon photodi-odes in CIS and indirect detection scheme with the pri-mary X-ray attenuators being scintillators. The directdetection is more suitable for X-ray energies up to about10 keV. The 1/ e attenuation length in silicon is 2.7, 17.5,127, and 962 µ m for 1, 5, 10, and 20 keV. Correspond-ingly, the fraction of X-ray attenuation and therefore thedetection efficiency decreases from 82.9%, 24.8%, 3.9%to 0.5% in a silicon pinned photodiode of thickness 5 µ m. At 20 keV, the 1 /e attenuation length in silicon ex-ceeds the 300 mm silicon wafer thickness of 775 µ m. Weshall mention without elaboration that other materialsand structures typically used in CIS such as the glass lidhave non-negligible effects on X-ray detection efficiencyfor energies below 20 keV.We consider the planar compact tiling configuration,D in Fig. 2, which is sufficient for X-ray energies below20 keV and plasmas with comparable or lower temper- atures. The direct detection model in silicon involvesX-ray to electron conversion, electron-hole (e-h) cloudpropagation, and noise model for the device. In siliconphotodiodes, 20 keV X-ray photoelectric (PE) absorption(91.6%) dominates over other processes such as Comptonscattering (3.1%) and coherent scattering (5.3%). ThePE fraction is more than 97% for X-ray energies less than10 keV. Based on the continuous slowing-down approx-imation (CSDA) and its modification at lower energies( <
10 keV), Fig. 3, the initial charge (e-h pairs) cloudproduced from the energetic electrons ( ≤
20 keV) gener-ated from PE process does not exceed 4.9 µ m, whichis comparable to the Vita 5000 CIS pitch of 4.8 µ m.The number of e-h pair created can be estimated as N eh = E X /E ± (cid:112) f E X /E for X-ray energy E X . E is3.64 eV, and Fano factor f is 0.13 for silicon. At E X = 5keV for example, N eh = 1374 ±
13. Further spread of thecharge cloud is due to e-h diffusion in silicon and chargesharing among multiple pixels . The read noise is 30 e − in the global shutter mode for Vita 5000 (dynamicrange of 53 dB for the full well depth of 13700 e − ). Weconclude that the resolution for direct detection of sin-gle X-ray photons is mainly determined by charge shar-ing among neighboring pixels, as confirmed by using avariable X-ray energy source (Amersham model: AMC2084), Fig. 3. FIG. 3. Photo-electron range in Si and LSO scintillator as afunction of X-ray energy based on CSDA model. Modifica-tion to CSDA model for Si is also included for E X <
10 keV.Experimental data using an variable energy X-ray source indi-cates that resolution for direct single X-ray photon detectionis mainly determined by charge sharing among neighboringpixels. The horizontal error bar corresponds to the energyspread of the X-ray source. The vertical error bar correspondsto 1 pixel width of 4.8 µ m of the Vita 5000 CIS. Next, we consider indirect detection schemes for10 keV and above energies, when X-rays first turninto a ‘cloud of visible photons’ by using a scintilla-tor. A few scintillators are summarized in Table. I.At 20 keV, the 1 /e X-ray attenuation lengths are 29.8,60.9, 91.6 µ m and 22.6 cm for Lu SiO (Ce) [LSO (Ce)],ZnO, (C H ) PPbBr [PPh4PbBr4] and plastic C H [EJ-228] scintillators respectively. Except for the plasticscintillator, the smaller 1 /e attenuation length than thatof silicon at E X = 20 keV may allow thin-film and 2Dstructures ( esp. for halide Perovskites) for efficient X-rayconversion, similar to the recent work on PAL . TABLE I. A comparison of light yield parameters of severalscintillators based on a modified Birks’s model, Eq. (1).
Scintillator S ρ k (ph/keV) (g/cm − ) ( µ m/keV)LSO(Ce) 30 7.4 3.2 × − ZnO 9.0 5.6 1.5 × − PPh4PbBr4 6-8 2.4 0.01 - 0.1EJ-228 10.2 1.0 0.13
The relative X-ray response for four different scintilla-tors has been measured using a Hamamatsu R2059 pho-tomultiplier tube (Bialkali 400S photocathode, quartzwindow, peak QE 27% at 390 nm) and the ArgonneAdvanced Photon Source (APS), Fig. 4. The plasticscintillator (EJ-228, 2.5 mm thick, emission peak 391nm), ZnO (0.3 mm thick, emission peak 380 nm) , LSO(3 mm thick, emission peak 420 nm), and PPh4PbBr4( ∼ est. at 400 nm). Theshape of the pulse is fitted with the function of the form I = I [exp( − t/t ) − exp( − t/t )] with t and t being therise and decay time respectively. FIG. 4. Characterization of the scintillator light yield anddecay time using the APS mono-energetic (29.2 keV, Sn K-edge) single-pulse X-ray in the hybrid mode. The rise timeand decay time, together with the relative light yield havebeen obtained from the pulse shape analysis.
The signals from individual X-ray photons can be es-timated as follows. We use the CSDA model to estimatethe initial size of the photon cloud generated by photo-electrons. Fig. 3 includes an example for LSO. The num-ber of photons emitted is 30 ph/ keV for 1 MeV photons in LSO. The photon yield decreases by a factor f y < f y ( E = 30 keV) = 0 .
85 and f y ( E = 10 keV) = 0 .
67 in LSO . At 29.2 keV X-rayphoton energy, the average number of photons emitted isabout 810. The critical angle is θ c = asin (1 /n ) = 0 . .Silicon has large optical refractive index ( n ) that is wave-length dependent. For example, n is 5.57, 4.65, 4.30, 4.08,3.79 at 400 nm, 452 nm, 500 nm, 550 nm, 689 nm respec-tively. Without AR coating, 30-40% of the incoming lightcould be lost at the silicon surface alone.The scintillator light yield ( L ν ) as a function of X-rayenergy uses a modified Birks’s model , dL ν dE = S k dEdx + k ( dEdx ) , (1)where S is the scintillation efficiency, dE/dx is the en-ergy loss of the particle per path length, and k is Birks’sconstant and material-dependent. The results here aresummarized in Fig. 5. The new ZnO and perovskite scin-tillator PPh4PbBr4 results are obtained through relativemeasurements shown in Fig. 4. The light-yield model andresults will be useful in further BiPC-X optimization. FIG. 5. Light yield model for X-ray energies from 10 to1000 keV, when the intrinsic light yield nonproportionalityis expected. The known values for LSO and EJ-228 are usedto obtain the new values for ZnO and PPh4PbBr4 based onrelative light intensities shown in Fig. 4.
IV. SENSITIVITY & RESOLUTION RESULTS
Single X-ray responses of different CIS models havebeen characterized using an Amersham variable energyX-ray source. Six pairs of K α and K β lines from Cu, Rb,Mo, Ag, Ba, Tb are excited by α particles from Amradioisotope. The lowest energy is Cu K α β FIG. 6. (1) An Amersham variable energy X-ray source usedfor the single photon sensitivity test. (2) The intensified imageof the X-ray source with Cu K α β α /K β . (4)-(6) Directsource images from Ag, Cu and Tb K α /K β X-rays.
Projection X-ray imaging using the direction detectionscheme was obtained using the APS synchrotron (ID 10),Fig. 7. Two Vita 5000 CIS were placed in a back-to-back stacked configuration along the X-ray beam path .The Fresnel numbers are 2.4 × and 1.8 × (1 mmspot size) for the front and back CIS respectively. Theresolution of 13 µ m is obtained in Fig. 8 from the line-out(y = 435) measurement of Fig. 7.In summary, we have shown that, due to the con-tinuing improvements in quantum efficiency, reductionin noise, and shrinking in pixel pitch, billion-pixel X-ray cameras (BiPC-X) are feasible based on commercialCMOS imaging sensors (CIS) and different tiling config-urations. A 2 × FIG. 7. X-ray images from a random wire pattern on twoback-to-back stacked CIS using the APS synchrotron. Thesmall rectangles are the spot size of the illumination.FIG. 8. Resolution test using a 20- µ m diameter gold-coatedtungsten wire. The FWHM for the wire projection (pixelnumber 280-285) is 2.7 pixels or 13 µ m. in detection efficiency. Possible applications of BiPC-Xinclude laser-produced and magnetically confined high-temperature plasmas when a few to 10s of keV X-raysare emitted to a wide field of view.We would like to thank Argonne APS ID10 staff, esp. John Katsoudas and Prof. Carlo Segre for help and co-ordination with scintillator measurements. The work issupported in part by the LANL Office of ExperimentalSciences (C3) program (contact: Dr. Bob Reinovsky).Z. W., supported in part by the LANL/LDRD pro-gram, also wishes to thank Drs. Blas Uberuaga, RichSheffield, Renyuan Zhu (Caltech), Liyuan Zhang (Cal-tech) for stimulating discussions and help. J. J. Ma, S. Masoodian, D. A. Starkey, and E. R. Fossum,
Optica , 1474 (2017). Z. Wang,
J. Instrum. , C12013 (2015). A. Dragone, C. Kenney, A. Lozinskaya, et al , J. Instrum. ,C11042 (2016). X. Li, P. Chu, et al. Nucl. Instrum. Meth. Phys. Res. A ,162414 (2019). E. Lee, M. R. James, K. M. Anagnost, et al , preprint atarXiv:2009.00555 (2020). D. J. Brady, M. E. Gehm, R. A. Stack, et al , Nature (2012)386-389. N. Teranishi, H. Watanabe, T. Ueda, and N. Sengoku,
Proc.IEDM (2012) 533. E. R. Fossum and D. B. Hondongwa,
IEEE J. Electr. Dev. Soc. (2014) 33. N. M. Winch, S. A. Watson and J. F. Hunter, Proc. SPIE , Medical Imaging 2017 101323U (2017). Z. Wang, et al. , J. Instrum. , C01035 (2018). C. Hu, et al. Nucl. Instrum. Meth. Phys. Res. A , 223 (2019). B. D. Rooney, J. D. Valentine and J. Li,
IEEE Trans. Nucl. Sci. (1998) 512. J. B. Birks,