visible detection and imaging in terms of sensitivity, resolution, power consumption, ease of use, size, weight, cost, etc. While cryogenically cooled detectors and arrays of detectors are relatively sensitive they are expensive and require cumbersome cooling liquids or power consuming thermoelectric coolers. The pyroelectric vidicon has made a great contribution toward the solution of these problems with the introduction of an uncooled ir image converter.1 However, it has insufficient sensitivity and spatial resolution for many applications as well as other disadvantages of a tube device such as size, weight, and power consumption. A technique has been developed to produce solid-state pyroelectric ir detector arrays and image converters. Predictions based upon known characteristics of CCD electronics,2-4 which would be used for signal readout, indicate that a noise equivalent temperature (NET) of 0.25°C or a noise equivalent power (NEP) of 2.5 × 10-10 w/Hz1/2 at 100-lp/cm spatial resolution for a 100 × 100-element array is possible.5 This sensitivity is equivalent to a single element cryogenically cooled detector, with mechanical scanning to produce an image, and is about a factor of 10 better than the pyroelectric vidicon at the same spatial resolution.
Pyroelectric detectors work by absorbing ir radiation and converting it into heat in a crystal. The changing temperature during heating or cooling of the crystal generates a displacement current between the detector electrodes. Construction of high resolution solid-state pyroelectric ir imaging devices has not previously been achieved because of heat conduction in the pyroelectric crystal. Thermal diffusion in the plane of the crystal during each frame reduces the spatial resolution, and diffusion through the crystal to the electronic substrate reduces the total temperature excursion and thus the sensitivity.
The method which has been developed to solve the above thermal problems is to chop the ir radiation at a relatively rapid rate to reduce thermal diffusion during each frame. The chopping rate is adjusted to match the thermal characteristics of the pyroelectric material and give the desired spatial resolution and substrate isolation. For low radiation intensity applications the signal from each detector element is then integrated over many frames in an electronic memory to recover much of the SNR lost by the fast chopping.
Using this technique, the means to connect the pyroelectric detector array to the substrate multiplexing electronics no longer presents a critical thermal problem. Heat does not
Fig. 1. (a) The 32-element linear pyrolectric detector array in 16-pin dip with SSFET multiplexing array; (b) the pyroelectric detector
array.
have sufficient time during one frame to flow through the detector and into the substrate. Thus connection of the two arrays can be done by well-developed technology such as flip-chip bonding or wire bonding rather than exotic thermally insulating methods.
A 32-element linear detector array has been fabricated using this technology. The array is assembled in a 16-pin integrated circuit dual-in-line package with a self-scanning FET multiplexing array as shown in Fig. 1 (a). The top array in the photo is the SSFET, and the lower one is the pyroelectric array which is attached by silver conductive epoxy to the header. The two arrays are connected by 32 ultrasonic wire bonds.
The pyroelectric array is composed of 50-μm thick LiTaO3 and consists of 32 elements on 4-mil centers. The elements measure 4 mils × 3.5 mils with 0.5-mil spacings. Gold connection electrodes are placed at each end of the array element to facilitate a variety of connection methods. A photograph of the pyroelectric array is shown in Fig. 1(b).
After coating the SSFET array with black paint to block visible radiation, the pyroelectric array was tested with a 1-mW He-Ne laser. This particular array is designed to work with pulsed radiation rather than chopped cw signals. The pyroelectric detector array generates both positive and negative signals during alternate half cycles of a radiation chopper, but the present multiplexing array reads out only one polarity of signal. Thus, after the readout of the first half frame, the pyroelectric signal biases the multiplexer below threshold, and no more signal is seen. With pulsed radiation the signal is read out after the pulse, and sufficient time is allowed before the next pulse for the array bias to return to equilibrium via diode leakage.
The performance of the array as a function of chopping frequency can be verified by varying the width of the radiation pulse. Spatial resolution, sensitivity, cross talk, and pulse repetition rates were measured as a function of pulse width and are discussed below.
Table I. Comparison of Calculated and Measured Detector Array Noise and Noise Equivalent Energy
Figure 2(a) is a plot of the responsivity of the detector array in V/μJ measured by increasing the pulse width from the He-Ne laser before readout. The response falls off above the 3-dB point of about 3.7 msec due to heat conduction into the substrate and into adjacent elements. The 3-dB time, due solely to the heat sinking effect of the substrate,5 is £3 dB = IIa2/k, where a is the detector thickness, and k is the thermal diffusivity. For the present array t3 dB is calculated to be 6 msec, which is 40% longer than the measured time. The contribution from lateral diffusion into adjacent regions of the crystal accounts for the difference.
The signal output of the array is measured by following it with a current mode amplifier. For calibration purposes a 10-pF capacitor was added to the feedback path to swamp out stray capacitance effects. The measured charge responsivity is then RQ (C/J) = Cf×Rv (V/J). The theoretical responsivity is RQ = αp(T)/pcpα, where α is absorptivity, p(T) is the pyroelectric coefficient, ρ is the density, and cp is the specific heat. The responsivity is calculated to be 5 × 10 - 7 C/J for 50-μm LiTaO.3 with α assumed to be 50%. The measured output voltage for a 2.5-msec pulse of the 1-mW laser was 0.12 V corresponding to a measured responsivity of RQ = 4.8 × 10 - 7 C/J, which is in excellent agreement with the calculated value.
The array sensitivity is defined as the noise divided by the responsivity. With self-scanned FET arrays there are two noise sources to consider. The charge transfer noise nCT arises from the uncertainty in resetting the capacitors to the same value during each multiplexing frame and is given by5
nCT = [(4/3)KTC]1/2,where K is Boltzmann's constant, T is absolute temperature, and C is the capacitance being reset. In the present array the capacitance being reset is about 20 pF, and so an rms noise of 3.3 × 10 - 1 6 C is calculated. A second noise source, fixed pattern noise, is generated by nonuniform leakage in the various elements of the detectors or the multiplexing electronics. This noise varies significantly from element to element (about 10% of the elements had 10 times the leakage of the average) and also as a function of the delay between multiplexing frames. Table I gives the calculated and measured noise charge and sensitivities for a worst case 5-msec delay between frames. For shorter delays the fixed pattern noise would be proportionately decreased.
The measured charge transfer noise is about 50% greater than calculated. Fixed pattern noise is considerably greater than the charge transfer noise but could be reduced by multiplexing the array just before the radiation pulse, storing the signal in a delay line, and then subtracting it from the multiplexed signal after the radiation pulse.
The dynamic range of the array is the ratio of the noise charge to the saturation charge. With a saturation charge of about 8 pC we have a dynamic range of about 104 with respect to charge transfer noise and 400 with respect to fixed pattern noise.
The cross talk, or the ratio of the signal on the two adjacent elements to the signal on the center element, is plotted in Fig. 2(b). Pulse widths shorter than 1.5 msec could not be reliably measured with the shutter setup being used. Nevertheless the minimum value shown of 8% is only slightly greater than the expected 6% electrical cross talk of the close spaced elements and connection electrodes. The pulse width t at which we would expect 15% thermal cross talk for a uniformily illuminated element is given by5 t = 1/(4IIn2k), where n is the number of lp/cm. The pulse width t is calculated to be 2.5 msec for the present array, in close agreement with the measurements of Fig. 2(b).
The maximum repetition rate for radiation pulses is a function of the output signal level since the readout charge must be replaced by diode leakage in the multiplexing array. Figure 3 shows curves of the measured output voltage vs the interval between successive radiation pulses. Shown is a curve for the element at the center of focus and one for the two elements on either side. The dashed curve is an extrapolation of the time delay required for full signal output corresponding to a saturation charge of 8 pC. These curves were measured with about 3-pF feedback capacitance. Thus, at 1-pA leakage current in the multiplexing array we expect about 2.4 sec to replace the 2.4 pC generated at an output of 0.8 V, in excellent agreement with the curve of Fig. 3.
Fig. 3. Detector array output voltage vs pulse interval.
The above experimental data correlate very well with our theoretical predictions of responsivity, noise, heat loading of the substrate, sensitivity, and thermal cross talk. These detector arrays can prove to be very useful as diagnostic devices on pulsed laser beams and other pulsed radiation sources. Other detector arrays with the elements elongated to 40 mils for spectrometer applications are planned. The technology used for array fabrication is amenable to low cost, high volume production. Newer arrays with 128 elements are at present under development. These arrays will use multiplexing electronic arrays that are capable of both positive and negative polarity readout. This will enable high speed chopping for cw signals and electronic integration to recover low level signals.
References 1. E. Stupp, Final Technical Report, contract DAAG53-75-C-0256,
for Night Vision Laboratory, USAECOM, Fort Belvoir, Va. (1976).
2. S. P. Emmons and D. D. Buss, J. Appl. Phys. 45, 5303 (1974). 3. A. M. Moshen, M. T. Tompsett, and C. H. Sequin, IEEE Trans.
Electron Devices ED-22, 209 (1975). 4. R. W. Brodersen and S. P. Emmons, IEEE J. Solid State Circuits
SC-11, 147(1976). 5. C. B. Roundy, Infrared Phys. to be published, 1979.
