Appendix A. MARGIE Segmented Scintillator and BiDirectional CCD Array

The 1.5 mm thickness of the MARGIE CdZnTe is optimized for energy resolution. This limits us, however, to a maximum energy ~200 keV. Thicker CdZnTe detectors could be used in order to go to higher energy, but the cost then goes up as well. We have used an alternate approach instead. In order to test the capabilities of high resolution segmented scintillators, we have used fine-grained arrays of "light pipes" of CsI scintillator viewed by silicon CCDs. Cherry et al.^1,2 and Gordon et al.³ have described the possibilities of growing very high resolution CsI microfiber arrays. We have demonstrated light yield comparable to that of bulk CsI (>30 photons/keV) and position resolution<100 µm. Fig. 1 shows an image of a tungsten mask (0.5 mm pixels) taken with a 1 mm thick CsI scintillator array produced at Radiation Monitoring Devices, Inc (RMD). The source is 40-50 keV X-rays from a dental X-ray machine; the scintillation light is viewed by a CCD.

Fig. 1. X-ray image of an etched tungsten mask obtained with a 1 mm thick CsI microfiber array and a CCD camera. The mask is a 71 x 73 URA pattern with 0.5 mm thickness and 0.5 mm minimum pixel dimensions.

MARGIE will extend its energy range to 600 keV and increase the scintillator thickness to 1 cm by relaxing the position resolution requirement to ~300 µm and forming the CsI scintillator columns in a metal matrix using a vacuum injection technique developed by RMD. A honeycombed metal mesh matrix is the basic structure. Such materials are used in aircraft wings and other applications requiring light weight and strength, and can be fabricated inexpensively in large areas with cell diameters as small as 300 µm and fill factors >85%. The thin metal walls between cells are opaque, and therefore eliminate the spreading of the optical light measured in a standard thick single scintillator and (to a lesser extent) in our microfiber arrays. A high atomic number metal limits the energy spread of the primary X-ray interaction due to Compton scattering and high energy photoelectrons. CsI is a suitable material for vacuum injection at reasonable temperatures. The scintillator material is injected into the mold in liquid form and then allowed to solidify into single crystals. Prototypes have now been produced with thicknesses up to 2 cm, and are currently being tested.

CsI (Tl) is reasonably transparent to its peak emission at 550 nm, and a silicon CCD can be used to detect the light with ~35% quantum efficiency. CCDs are, for many applications, nearly ideal light detectors. They are reliable, and provide excellent position resolution with low noise. For our application, however, a standard CCD is slow: CCDs typically operate at no better than video rates (20 - 30 frames/s). For studies of pulsars (e.g., the Crab), this is too slow, for example. It is also too slow to operate in conjunction with an active charged particle anticoincidence shield. (In the case of MARGIE III, a cosmic ray proton will strike the shields or the mask and potentially generate a bremsstrahlung photon approximately every 250 µs. With 30 µs CCD timing, individual cosmic ray shield hits can be flagged and then removed during the analysis.) We have therefore developed a new CCD readout architecture providing 30 µs time resolution and lower noise performance. This BiDirectional Fast Timing CCD² will have applications to medical imaging, inspection on moving assembly lines, reconnaisance from fast moving vehicles and aircraft, and CCD spectroscopy (where sufficient timing resolution is necessary to insure only one photon "hit" per pixel per readout).

With a standard CCD readout architecture, the photosites collect light for some integration time t_int >> 50 ms. If a photon strikes a photosite (i,j) during the integration, a photoelectron is generated at that pixel. At the end of the integration, the charge on each pixel is clocked vertically upward one row at a time. After each vertical charge transfer, one row of charge has been transferred into the horizontal readout register at the top of the chip. Now all the individual pixel (column) charge packets in the horizontal readout register can be clocked horizontally to the output amplifier. Then the entire array is again shifted vertically; the next row is shifted horizontally; and the charge transfer continues until all rows and columns have been shifted to the output amplifier.

Rather than the photon arriving at pixel (i,j) during the integration time, however, imagine that the photon arrives at (i,j) at a time t_o after the start of the readout. If the time between vertical transfers is t_row, then t_o / t_row rows will have been clocked upwards by the time the photon arrives at (i,j). A photoelectron will still be produced at (i,j), but it will be read out as if it had been produced during the integration time at pixel (i + t_o / t_row, j). In the standard situation, a photon arriving during the readout and producing a spurious signal would be considered a background. A shutter is frequently used to block off the CCD during the readout in order to prevent this background.

In the case of low fluxes, however, it is possible to use this situation to determine the arrival time t_o. Imagine that X-rays are absorbed in a scintillator microfiber or metal matrix array with position resolution 100µm, and the scintillator is viewed by a Bi-Directional CCD array with pixel pitch 50 µm. The optical signal from the scintillator will then be seen simultaneously by pixels in at least two contiguous columns. If columns are clocked continuously (i.e., if t_int = 0), with odd columns clocked upward and even columns clocked downward, then the signature of an event will be hits in two adjacent columns at (i + t_o / t_row, 2j + 1) and (i - t_o / t_row, 2j + 2). The actual row position is the average [(i + t_o/t_row) + (i - t_o/t_row)]/2, and the event arrival time is given by the difference t_row [(i + t_o/t_row) - (i - t_o/t_row)]/2.

If there are n_rows rows, then the time to read out the entire array is n_rows t_row ( = 15 ms for the case of a 512 x 512 array clocking vertically every 30 µs). The flux must be sufficiently low that the chances of seeing more than one event per two adjacent columns per full-frame readout time is small:

F<[(2N²_rowsd²t_rows)]^-1

where d is the pixel size. (For d = 50µm in this example, the limiting flux is then 3 x 10³ X-rays cm^-2 s^-1.) For a standard 50 ms readout, the train of 5 MHz horizontal clock pulses generates noise and draws power, as does the ADC (which must perform a pulse height conversion in 380 ns). In the Bi-Directional approach, however, note that at the limiting flux, the average number of hits per row is 1/2n_row << 1.

Proper readout chain design is critical to low-noise CCD performance. For the BiDirectional CCD, with an independent amplifier chain for each column, the readout circuitry will be implemented in 1.2 µm CMOS directly on the CCD. The design is currently being developed by Lawrence Berkeley Lab and Suni Imaging Microsystems, Inc., based closely on the previous LBL designs for the SVX2 chip used at Fermilab^4,5 and the silicon detector on the ACE spacecraft cosmic ray heavy ion experiment. Two readouts will service each CCD, with a total footprint (2.6 cm)². Each readout ASIC will have 256 input channels (i.e., 256 for the top, 256 for the bottom of the CCD). Each readout channel will have an independent preamplifier, amplifier, double correlated sampling noise suppression, sample and hold, and ADC with a 30 µs event clocking rate. The ADC uses a low-power Wilkinson converter with a Gray code counter sensitive to both rising and falling edges of a common clock measuring time-above-threshold of a single decreasing ramp for all 256 channels. ADC resolution will be 9 bits, with a power dissipation ~2 mW/channel for the full readout chain.

A BiDirectional CCD has been designed at Suni, and is currently being fabricated at Orbit Semiconductor, a leading CCD foundry. The first prototypes are due off the fabrication line in June 1996, and will be DC-tested, packaged, and delivered to LSU in July.

Bibliography

1. M.L. Cherry et al., Proc. 24th Intl. Cosmic Ray Conference, Rome 2, 45 (1995).
2. M.L. Cherry et al., MASS/Airwatch Workshop Report, Huntsville, p. 259 (1995b).
3. J. Gordon et al., in X-ray and Ultraviolet Sensors and Appl., SPIE Proc. 2519, 2 (1995).
4. S. Kleinfelder et al., IEEE Trans. Nucl. Sci. 35, 171 (1988).
5. T. Zimmerman et al., IEEE Trans. Nucl. Sci. 42, 803 (1995).