ISSN 1063-7397, Russian Microelectronics, 2008, Vol. 37, No. 2, pp. 114–120. © Pleiades Publishing, Ltd., 2008.
Original Russian Text © I.I. Lee, 2008, published in Mikroelektronika, 2008, Vol. 37, No. 2, pp. 131–138.
With infrared focal-plane arrays (IRFPAs), the best
way to achieve high sensitivity is still to employ cooled
hybrid models, although microbolometer-based arrays
did make good progress in recent years.
The designers of hybrid IRFPAs have been success-
ful in dealing with challenges posed by second-genera-
tion systems and, to an extent, by third-generation ones.
In particular, two accomplishments appear to be the
(1) Major issues are resolved relating to the materi-
als for and the fabrication processes of IRFPAs [1, 2].
(2) Silicon readout circuits are developed with
dynamic capabilities such as readout windowing and
variable integration time and frame rate [3, 4].
As a result, one-dimensional (1D) arrays, including
ones with time-delay integration, and two-dimensional
(2D) arrays are now available for operation in the wave-
length range 2–12
m, with pixel counts exceeding
At the same time, the sensitivity of IRFPAs is
known to decrease at 8–12
m due to a limited charge-
handling capacity of readout circuitry .
By way of illustration, Fig. 1 shows the noise-equiv-
alent temperature difference (NETD) of thermal imag-
ers calculated as a function of wavelength and readout
charge capacity . Curves
refer to the ideal
case where the integration time and the frame time are
the same, 20 ms. Under these conditions, the NETD
decreases with increasing wavelength, as with a perfect
thermal imager . Here, a thermal imager is described
as perfect if its detectivity is close to that in the back-
ground-limited regime, the readout circuit has an ade-
quate charge-handling capacity to utilize an incoming
photosignal completely throughout the frame time, and
the noise level of the readout circuit is lower than the
correspond to charge-handling capaci-
tively, which determine the integration time. Notice
that the NETD starts growing with wavelength once
this has passed a value as small as 3.2
m. The integra-
tion time is 30 to 100
s or shorter for the rightmost
point of each curve, being less than the row readout
time. In this case, 2D arrays offer no advantage over 1D
ones in terms of NETD.
in Fig. 1 is an experimental result for an
InSb IRFPA operating over the wavelength range 3–5.4
. It gives an NETD of 5.5 mK, which makes up about
50% of that for the perfect thermal imager.
in Fig. 1 is an experimental result for a
HgCdTe IRFPA operating at 1–10.3
m . The NETD
of this system is about 13 mK, a mere 5% of that for the
perfect thermal imager.
A required charge-handling capacity increases with
desired cutoff wavelength. With operating wavelengths
m, it is sufﬁcient to have a charge-handling
capacity in the range (2–10)
to the wavelength range 8–14
m necessitates raising
the minimum capacity to
electrons if we are to
achieve a high standard of sensitivity. Otherwise, a
charge-handling capacity of, e.g.,
would correspond to an integration time as short as a
few percent of the frame readout time; in other words,
the integration time would be of order the row readout
time at the very best. This is what prevents 2D IRFPAs
for the longer wavelengths from attaining a sensitivity
better than that of 1D IRFPAs with four to eight time-
It is held that improving silicon readout circuitry is
crucial to further progress in IRFPAs for the wave-
length range 8–12
m. Speciﬁcally, two challenges are
to be met:
Silicon Readout ICs for Third-Generation 2D IR Focal-Plane
Arrays Operating over the Wavelength Range 8–12
I. I. Lee
Institute of Semiconductor Physics, Siberian Branch, Russian Academy of Sciences, Novosibirsk, Russia
Received April 18, 2007
—A new design is described of silicon readout IC for 2D long-wavelength hybrid IR focal-plane
arrays. Its essential feature is that the input array is made up of 2
2 subarrays of cells. This provides an almost
tenfold increase in charge-handling capacity, a major parameter of the readout ICs considered. The design also
includes an adaptive digital preprocessing unit in on-chip form.
PACS numbers: 85.40.-e