A low-noise silicon-based 20μm*20μm uncooled thermoelectric

infrared detector

Mohammad J. Modarres-Zadeh, and Reza Abdolvand

Dynamic Microsystems Laboratory, Oklahoma State University/Tulsa,

103 HRC, 700 N. Greenwood Ave. Tulsa, OK, 74106

Abstract

Presented is an uncooled surface-micromachined thermoelectric (TE) infrared detector that features P-doped and N-

doped polysilicon wires as the thermocouple pair and an umbrella like optical cavity as the absorber to achieve a

high fill factor. A responsivity as high as1800V/W @5Hz and a response time of smaller than ~10ms are measured

in vacuum when viewing a 500K blackbody with no concentrating optics at room temperature. The reported

responsivity is more than 10 times higher than the value reported earlier [1] from similar structures due to the

improvement in the thermoelectric coefficient and the thermal isolation of the cell. Finite Element Analysis is used

to predict the detector’s performance and the results are in a good agreement with the measurements. The dominant

source of noise is also investigated in these thermoelectric IR detectors and it is believed to be Johnson noise when

they are operated under an open circuit condition. The fabricated detectors have resistances in the range of 20 to

70KOhm resulting in a Johnson noise of about 20 to 36 nV/Hz^0.5. The specific detectivity (D*) is calculated to be

higher than 10^8cmHz^0.5/W. To the best of our knowledge, this is the highest reported D* for such small

thermoelectric IR sensors. The measured NETD is 120mK with an f/1.5 lens.

Keywords: Thermoelectric, uncooled Infrared detector, high responsivity

1. Introduction

Infrared is described as the electromagnetic radiation at wavelengths longer than visible light. Any material above

zero Kelvin emits infrared radiation and since the radiated power is a function of the material temperature, the

detected radiation can be used to estimate the material temperature. This is the most basic application of IR

detectors.

In any application, an electric signal proportional to the incident radiation must be generated. Today, several sensing

principles are utilized to detect IR radiation. For example, cooled photon detectors are very sensitive and offer a fast

response resulting in a very high quality image. A good review on the progress of such detectors can be found in [2].

Because of high manufacturing costs and requirements on operation at low temperatures (sometimes as low as few

Kelvin [2]), an alternative to such technology has been always demanded that is commonly referred to as uncooled

IR detection.

To date, the most widely available uncooled IR detectors are resistive bolometers, which operate based on the

change in the resistance of an irradiated material and the output electric signal is generated by biasing this material.

The performance of all uncooled infrared sensors is mostly limited by noise. Out of different sources of noise,

Johnson noise is believed to be dominant [3]. However, with recent developments and reduction in pixel size 1/f

noise and the temperature fluctuation noise (

where C is the heat capacity of the detector) have become

significant [3].

1/f noise can be eliminated in Pyroelectric detectors. Pyroelectricity is a property of some materials to momentarily

accumulate surface electric charges in response to temperature changes. Therefore, a mechanical chopper is required

for operation of pyroelectric devices. Without mechanical choppers, the surface charge and consequently the output

voltage cannot be sustained for constant incoming IR radiation. Although using a chopper could effectively

jeff.modarres_zadeh@okstate.edu; phone 1 918 594-8642; http://microsystems.okstate.edu/

Infrared Technology and Applications XXXIX, edited by Bjørn F. Andresen, Gabor F. Fulop, Charles M. Hanson, Paul R. Norton, Proc. of SPIE Vol. 8704, 87041R · © 2013 SPIE

CCC code: 0277-786X/13/$18 · doi: 10.1117/12.2016283

Proc. of SPIE Vol. 8704 87041R-1

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/05/2013 Terms of Use: http://spiedl.org/terms

eliminate 1/f noise, long term drift , and offset variation between the elements in an array, but in many applications

it is an undesirable component of a system [4].

On the other hand, thermoelectric (TE) detectors neither require choppers nor are limited by the 1/f noise. In these

detectors, a TE material is placed between a thermally isolated structure and the surroundings (substrate). When the

suspended structure is irradiated, its temperature rises slightly above the substrate. Thus, TE material generates a

signal which is proportional to the temperature difference. The readout circuit for TE detectors is commonly

designed with high input impedance and only allow negligible current (less than 10pA for many op-amps) to be

drawn from the detector resulting in negligible 1/f and shot noise (

√ where is the average current flow

[5]).

The main disadvantage of TE detectors is their low responsivity due to their inefficiency in converting heat to an

electric signal. With new developments in fabrication and characterization of nanostructures, it has been shown that

the TE efficiency of nano-engineered material such as silicon nanowires is substantially higher [6]. We propose that

with the pixel size becoming smaller, nanowires could be implemented as both the TE wires and structural

suspension in the detector structure resulting in enhanced performance.

In this work, we report on fabrication and testing of uncooled TE detectors using thin polysilicon (~115nm) as both

TE and structural tethering material with relatively high responsivity and low noise at a small pixel size of 20 um by

20 um. In the following, the detector design and its theory of operation will be explained followed by the fabrication

process. At the end, test setup and measurement results will be presented.

2. Theory of operation

It is well known that for maximizing the thermoelectric efficiency in power generation applications, the figure of

merit, “Z”, should be maximized. Z is defined as

where S is Seebeck coefficient, is electrical resistivity, and K

is thermal conductivity and all these parameters are functions of temperature. However, when TE materials are

utilized as a sensing element, Z may no longer be the best figure of merit, since the delivered power to the sensor

load (for example the readout op-amp) is ideally zero. Hence, we intend to redefine an appropriate figure of merit.

In infrared sensors, D* or specific detectivity is defined as √

(where is the RMS noise voltage, is the

responsivity, and is the area of the detector). With this definition, the detectivity can be utilized to compare the

performance of different detectors and/or one detector with varying parameters [5]. If Johnson noise is considered as

the dominant source of noise, the noise voltage amplitude is √ where is Boltzmann’s constant, T is

temperature in Kelvin and R is the detector resistance (

). DC responsivity for TE sensors is calculated from

where N is the number of thermocouples, S is Seebeck, is the optical absorption and G is the thermal

conductance from the suspended island to the substrate. This can be written as where k is the average thermal

conductivity of the thermocouple. If the above equations are substituted in the equation for D* and after cancelling

out the spatial/frequency parameters we can observe that

√ . Therefore,

√ is the relevant ratio that has to

be maximized (compare to

). This result confirms that Z is still a reliable figure of merit that can be used to

compare efficiency of different TE material in infrared sensing application. However, for more accurate comparison,

√ is preferred. Since Z has been reported for many materials, we will be using it as well.

To maximize ZT (Z * T where T is the temperature in Kelvin), a material with a large Seebeck coefficient, a low

resistivity and a low thermal conductivity is required. None of the materials that are readily available in CMOS

fabrication process offer high ZT if CMOS compatibility is of interest. For instance, ZT of bulk silicon is 0.01 at

room temperature [6]. As mentioned earlier, recent studies suggest that silicon nanowires exhibit substantially

higher ZT compared to the bulk value [6]. This is mainly because in nanowires the lattice part of the thermal

conductivity can be greatly reduced by boundary scattering without significantly affecting the Seebeck coefficient or

the electrical conductivity.

Since thermal conductivity can be independently controlled by the physical dimensions of the TE wires, a new

figure of merit called power factor and defined as

can represent the efficiency of TE wires. In the case of

semiconducting TE wires the electrical resistivity (ρ) is reduced by doping. However, doping generally has an

Proc. of SPIE Vol. 8704 87041R-2

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/05/2013 Terms of Use: http://spiedl.org/terms

293.7

293.6

293.5

293.4

293.3

293.2

adverse effect on the Seebeck coefficient lowering the responsivity. Nevertheless, maximum power factor is shown

to be achieved in heavily doped semiconductors [7]. Instead of silicon, polysilicon is used as the thermoelectric

material in this work due to the ease of deposition in low pressure chemical vapor deposition (LPCVD) furnaces and

the excellent control over the desired thickness of the film. The polysilicon film in this work is doped to a

concentration level of - (measured by a four point probe instrument).

The detector structure is presented in Figure 1(a). Finite element method (FEM) is utilized to predict the detector’s

performance. The detector is comprised of an absorber on top, an absorber post made of the same material as the

absorber and connected to the suspended membrane, and the TE wires (shown in red). The TE wires serve also as

supports to suspend the structure. When the absorber is irradiated, the rise in its temperature (shown in Figure 1(b))

causes the suspended membrane temperature to increase and because of the relatively high thermal conductance

between the absorber and the suspended membrane compared to the thermal conductance between TE wires and

substrate, it can be assumed that suspended membrane and the absorber are at the same temperature. Details and

discussion on absorber structure can be found in [1]. It is worthwhile to evaluate the detector performance as the

design parameters change. A complete list of the physical dimensions and variables are presented in Table 1.

According to Wien’s displacement law, human body has a maximum radiation at around 10um wavelength. If the

area of a human body is assumed to be , by using Planck’s radiation equation, the total transmitted power in the

range of 8-14 um is about 170 watts. If the body is radiating isotropically and is a meter away from the detector,

is the power density incident on the detector. This value is used for the absorber heat source.

Table 1. Parameters used in the simulation.

In this simulation, the effect of thermal radiation off of the absorber is also included. In COMSOL, Stefan-

Boltzmann equation is used to take the radiation losses into account. However, in this way, all the radiated power in

all wavelengths are combined which results in a number much larger than the actual value since the radiation losses

should only be taken into account in the range that absorption is significant. Assuming that the absorber temperature

is not deviating from its initial value by more than 2 to 3 degree, a correction coefficient is calculated from Planck’s

radiation equation and is multiplied to the emissivity of absorber. For an absorber with the initial temperature of

293.15 K, the correction factor is for 8-14 um window.

Parameter Value

Pixel size 20 um * 20 um

Suspended membrane size 5 um * 5um

Suspended membrane

thickness

300 nm

Absorber post height 4 um

Absorber post bottom radius 2.2 um

Absorber post inclining angle 20 degree

Parameter Value

Absorber thickness 1 um

TE Width Variable 100 to 1200

nm

TE thickness 115nm

Absorber absorption 80% - Flat 8-14um

Polysilicon thermal

conductivity

Variable from 1 to 30

W/m.K

Absorber Absorber post

Suspended

membrane TE wires

Figure 1(a). The detector structure, (b) The simulated and color coded temperature distribution on

the detector.

Proc. of SPIE Vol. 8704 87041R-3

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/05/2013 Terms of Use: http://spiedl.org/terms

The steady state responsivity as described above is proportional to the thermal conductivity (

).

However when radiation losses are taken into account, the detector performance deviates from this simple

proportional equation especially when TE wire thermal conductance is small and ultimately approaching zero.

Since the output voltage of the detector is for a constant radiation, the

temperature of the suspended membrane is a good indicator of how different parameters are affecting the output

signal. For instance, if TE thermal conductivity could be independently reduced by nano-structuring, roughening the

surface, and/or introducing nano-holes, the suspended membrane temperature would be higher which translates into

a larger output signal or higher responsivity. For a detector with TE width of 200 nm and a constant substrate

temperature, the suspended membrane variation with thermal conductivity is shown in Figure 2 (left). Figure 2

(right) shows the error introduced when absorber radiation losses are ignored.

If the material thermal conductivity is presumed to be constant, then it is useful to know the variation of the

suspended membrane temperature versus the widths of the TE wires. Thickness of the TE wires is controlled at the

time of the deposition and their widths can be adjusted by lithography. The simulated increase of T across the TE

wires presented in Figure 3 is based on an assumption that thermal conductivity of polysilicon is 5 W/m.K.

Figure 2. Left plot shows the variation of the suspended membrane temperature with respect to substrate vs. thermal

conductivity of TE wires when radiation losses are included and ignored. Right plot shows the percentage of the error

introduced when radiation losses are ignored as a function of the temperature difference across the TE wires.

0

0.5

1

1.5

2

2.5

3

0 5 10 15 20 25 30

ΔT

(K

)

Thermal conducitivity of polysilicon

(W/m.K)

Including radiation losses

Ignoring radiation losses

1

10

100

0 1 2 3

Err

or

(%)

ΔT (K)

0

0.5

1

1.5

0 200 400 600 800 1000 1200

ΔT

(K

)

TE wire width (nm)

Figure 3. Temperature variation of the suspended membrane with respect to its substrate vs. the widths of the TE wires.

Proc. of SPIE Vol. 8704 87041R-4

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/05/2013 Terms of Use: http://spiedl.org/terms

3. Fabrication

Fabrication of the devices described in the previous chapter requires 6 different masks. The process starts with a

blank silicon wafer. A thin (~ PECVD (plasma enhanced chemical vapor deposition) silicon dioxide film is

first deposited and patterned to form small islands where each island will serve as a sacrificial layer for a detector

structure. Later in the process, these islands will be removed in Hydrofluoric Acid (HF) to leave the detectors

suspended. Next, a low-stress silicon nitride film is deposited in an LPCVD (Low pressure CVD) furnace at

850 , and patterned using photoresist mask in an inductively coupled plasma (ICP) tool (Figure 4(a)). Then,

in-situ doped P-type polysilicon film is deposited in an LPCVD furnace at a temperature about

followed by patterning using an oxide mask. This mask will be kept for the next step which is P-type polysilicon

deposition. The oxide mask will prevent counter dopants to diffuse to the patterned N-type polysilicon layer and

also restrains the dopants in the N-type polysilicon from out diffusion. Annealing at 950 for 30 minutes in

nitrogen is required for in-situ doped films to fully activate the dopants. The N-type polysilicon film is then

patterned using a photoresist mask. At this point, the width of the two patterned polysilicon wires can be adjusted by

timing the isotropic etching of polysilicon in SF6 plasma (Figure 4(b)). The etching recipe is optimized to have

minimal effect on the oxide and nitride layers that already exist on the wafer.

Figure 4. The simplified schematic of the fabrication process flow.

Metal contacts are deposited at this step on the polysilicon layers. However, since the first polysilicon wire is still

covered with an oxide mask, the oxide will be locally removed at the spot where the metal contacts are sputtered. As

discussed in [1], a parylene layer can be optionally deposited and patterned at this point. This layer could be used to

support the thin and fragile TE wires. Deposition of parylene thin film is carried out in SCS 2010 followed by

patterning in Oxygen plasma. It will be shown that this supporting layer affects the performance of the detectors in

spite of its excellent low thermal conductivity. Thus, after releasing the detectors, the layer could be completely

removed in oxygen plasma. Next, the second sacrificial layer of PECVD silicon dioxide (4 um) is deposited. The

device absorber which will be deposited on top of this layer has to be anchored to the nitride island. In order to do

so, a circular pattern is etched in the oxide following by deposition of an absorber film stack consisting of a 1 um

silicon nitride layer sandwiched between two nichrome layers (Figure 4(c)). Patterning the absorber layers is a

combination of wet and dry etching. Top nichrome film is etched in nichrome TFN etchant. This etchant has low

etch rate on the underneath silicon nitride. Dry etching of PECVD Silicon nitride is carried out in ICP. The etch stop

for this dry etching is the underneath nichrome layer, since nichrome is not affected by the nitride etching process

and the nitride layer can be over etched. After that, the last layer of the absorber is etched in nichrome TFN. All the

wet etching processes are timed to minimize undesired undercuts. At the very end, the detector is released by

Silicon Silicon dioxide Silicon Nitride TE wires Absorber layer

(a) (b)

(c) (d)

Proc. of SPIE Vol. 8704 87041R-5

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/05/2013 Terms of Use: http://spiedl.org/terms

removing the sacrificial layer in hydrofluoric (HF) acid solution (Figure 4(d)). A scanning electron microscope

(SEM) image of a completely fabricated device is shown in Figure 5.

4. Test setup and results

The schematic of the test setup for measuring the responsivity is presented in Figure 6. To measure responsivity, a

NIST calibrated sensor is placed inside a vacuum chamber. Then an appropriate infrared window (Germanium with

antireflection coating for 8-14um) is placed on top of the chamber. A blackbody with the temperature set to 500 K is

located a meter away from the chamber and a chopper is placed between the window and blackbody as shown in

Figure 6. At first, the power density of the incoming IR radiation is measured using a NIST calibrated sensor. Then

the sensor is exchanged with a fabricated detector and the

output voltage of the detector is measured utilizing

chopper stabilization techniques.

As described in [1], parylene is a chemically stable and

inert polymer that can be deposited at room temperature

and because of its low thermal conductivity, it is a great

candidate to be used as a supporting layer for thin fragile

TE wires. A wafer was fabricated with 400nm parylene

following the procedure outlined in the fabrication section.

The output response of ten detectors with different designs

was measured. Then, the parylene layer was removed

using oxygen plasma followed by testing the same

detectors under the same condition. The test results are

summarized in Table 2.

As it can be seen, depending on the detector structures, the

improvement in the detectors’ performance is different but on average, a 2.5X improvement is achieved by

removing the parylene layer. This result shows that the heat conduction through the parylene layer is comparable to

that of TE wires.

By using a chopper, the detector time constant can also be measured. One method for measuring time constant is to

record the rise or fall time of the detector’s amplified output signal. However, this technique is not very accurate

since the window on top of the vacuum chamber and the chopper opening are producing a convolution in space,

which causes the measured value to be greater than the actual value. A more accurate method is to increase the

chopping frequency and plot the responsivity versus frequency and match that with the responsivity equation. The

matching is the thermal time constant of the detector. Figure 7 shows the measured data and a fitted line using

responsivity equation with of 8ms.

Figure 5. The SEM of a fabricated detector.

Absorber

TE beam

Metal contact

Figure 6. The measurement setup

Black body

Vac. Pump Detector Chamber

Ge Window

Chopper

Proc. of SPIE Vol. 8704 87041R-6

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/05/2013 Terms of Use: http://spiedl.org/terms

Next, the detector noise is measured. The detector is placed inside the vacuum chamber and after probing the device,

the chamber opening is covered by an IR opaque material. A Signal recovery-7270 lock in amplifier is used to

amplify the signal and a Tektronix (TDS5104B) digital oscilloscope is used to display the amplified signal and to

perform the measurements. All the electronic couplings are set to DC and no line filter is used. Only a single-pole

low pass filter inside the oscilloscope probe with of 100 Hz is used. The input of the amplifier is first shorted

and the noise associated with the electronics is measured by taking samples for about 10 minutes/5K samples. Then,

noise of a metal film resistor with a value equal to the detector resistance is measured. At the end, the noise of the

detector is back calculated by measuring samples from the detector output itself. For a detector with 47 KOhm

resistance a thermal noise of

√ is expected. However, assuming that the measured noise is distributed evenly in

frequency, about

√ higher noise is recorded. Since the device has not been wire-bonded and is probed by signal

probes, authors believe that the excess noise comes from the vibration of the probes on the detector pads plus the

vibration induced from the chamber vacuum pump. Performance of a typical detector is summarized in Table 3.

Table 2. The responsivity of several detectors before and after removing 400nm thin-film parylene off of the TE arms.

Device number

Output voltage before

removing the parylene

(uV)

Output voltage after

removing the parylene

(uV)

1 4.15 10.9

2 3.6 14.2

3 10.9 24.1

4 6.65 13.4

5 4 12.6

6 9 16.7

7 5 15.5

8 3.5 15.2

9 15 22.4

10 9.7 22.4

Table 3. The Performance of a typical detector with no concentrating optics when viewing a 500 K blackbody.

Resistance

KOhm

Responsivity

Noise

√

Specific Detectivity

√

45 ~ 1800 ~

Figure 7. Responsivity vs. ω – The blue marks are the measured data and the red line is the TE

responsivity model for τ=8ms

0.45

0.55

0.65

0.75

0.85

0.95

1.05

1 10 100 1000

Re

spo

nsi

vity

ω (rad/s)

Model (τ = 8ms)

measured

Proc. of SPIE Vol. 8704 87041R-7

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/05/2013 Terms of Use: http://spiedl.org/terms

5. Conclusion

An uncooled surface-micromachined thermoelectric (TE) infrared detector was designed, fabricated and

characterized that features P-doped /N-doped polysilicon as the thermocouple pair and an umbrella like optical

cavity as the absorber to achieve fill factor as high as 90%.

Theory of the operation along with some design considerations has been discussed. FEM analysis is used to predict

the device performance when different parameters change. It has been shown that radiation losses can become

important when the detector temperature rises beyond 1 - 2 Kelvin above substrate temperature.

A new figure of merit (

√ ) is defined for more accurate comparison of different TE materials for sensor

applications.

Fabrication steps have been briefly explained along with the measurement setup. At room temperature, a

responsivity of 1800

at 5Hz and the response time of less than 10ms are measured in vacuum when viewing a

500K blackbody with no concentrating optics. The very small reduction in the thermal conductivity of thin-film

polysilicon TE wires is believed to be a key in achieving such a large responsivity. The detector noise is measured

by recording its output, and shown that it is slightly above detector thermal noise. The D* is calculated to be about

√

which to the best of our knowledge is the highest reported D* for 20um thermoelectric devices. The

measured NETD is 120mK with f/1.5 lens when the object temperature is set to 500 K.

6. Acknowledgments

This work was generously supported by Amethyst Research Incorporated through funds received from the

Oklahoma EDGE program.

REFERENCES

[1] Modarres-Zadeh M. J., Carpenter Z. S., Rockley M. G., and Abdolvand R., “Parylene supported 20um*20um

uncooled thermoelectric infrared detector with high fill factor,” Proc. SPIE 8353, (2012).

[2] Rogalski, A., “Infrared detectors: Status and trends,” Prog. Quant. Electron., 27(2-3), 59-210 (2003).

[3] Li. C., Skidmore G. D., Howard C., Han C. J, Wood L., Peysha D., Williams E., Trujillo C., Emmett J., Robas

G., Jardine D., Wan C. F., and Clarke E., “Recent development of ultra small pixel uncooled focal plane arrays at

DRS,”. Proc. SPIE 6542, 1Y.1-1Y.12(2007).

[4] Rogalski A. , [Infrared detectors], Taylor and Francis Group, Boca Raton Florida, 697(2011).

[5] Daniels A., [Infrared Systems, Detectors, and FPAs], SPIE, Bellingham, 67-76(2010).

[6] Vining C. B., “Materials science: Desperately seeking silicon,” Nature, 451-132 (2008).

[7] Snyder G. J., and Toberer E. S., "Complex Thermoelectric Materials," Nature Materials 7, 105-114,(2008).

Proc. of SPIE Vol. 8704 87041R-8

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 09/05/2013 Terms of Use: http://spiedl.org/terms