Project Report Spectrometer on A Chip | ECE 445, Study Guides, Projects, Research of Electrical and Electronics Engineering

Download Project Report Spectrometer on A Chip | ECE 445 and more Study Guides, Projects, Research Electrical and Electronics Engineering in PDF only on Docsity!

SPECTROMETER ON A CHIP

by Seth Robert Bank Jeffrey Scott Kline ECE 345 T.A. Ben Floering 5/4/ Project #

ABSTRACT (SRB)

Over the years spectroscopy has proven to be one of the best methods of determining the chemical makeup of a sample. However, spectrometers have typically been large and not portable. Utilizing silicon based fabrication technology and unique device geometry, it is possible to shrink the dimensions of the device while at the same time increasing the theoretical limit of resolution to as large as desired. This paper describes the detailed design of a prototype spectrometer that uses a prism for refraction and p-i-n photodetectors for detection. The design, with modifications that are also discussed, can potentially improve the resolution currently available with transmission grating spectrometers. Also, the testing of a fabricated p-i-n photodetector is detailed and the results presented. Finally, the paper discusses potential improvements in the design needed to increase wavelength resolution to commercial standards. ii

1. INTRODUCTION (SRB)

Knowledge of a material's spectrum is useful for determining its chemical makeup. A handheld device capable of measuring the spectrum of a light source would have many applications. One such use is as a field unit for analyzing water supplies for various contaminants and bacteria. Before continuing further, it is necessary to acknowledge that this paper will be slightly different from norm. Initially, it will detail the design of a spectrometer on a chip. Unfortunately, due to time constraints, it was impossible to complete a working prototype. Consequently, the testing will be of a p- i-n photodetector array and external circuit designed for use in the spectrometer. Since the design and fabrication of the photodetectors was initially a subset of the spectrometer, this method of presentation will fully develop the theory behind the tested device. Additionally, it will have the benefit of showing the way for the eventual construction of a working spectrometer. The spectrometer will use a prism to separate the incident light into its spectrum via refraction. The separated light will fall on a one-dimensional array of photodetectors fabricated on bare silicon. Their electrical state (on or off) will be displayed, through some external circuitry, on an array of LEDs corresponding to a certain wavelength range of the incident light as illustrated in Fig. 1 (All figures appear at the end of the report.) The p-i-n photodetectors are in a unique planer configuration consisting of rows of diffused p and n wells sharing a common intrinsic region (See Figure 2.) This is made possible by the wide blocking wells that provide device isolation, allowing the array to resolve the continuous spectrum. The design breaks into 5 interdependent phases: optics, design and layout of the photodetector, fabrication, the external circuitry, and finally combining the first four into the final product (See Figure 3.) The first phase is the optics. Referring back to Fig. 1, it is seen that the concepts behind this phase are quite simple. The incident light source travels through an aperture on one face of the prism and is refracted. The refracted light then travels through the prism and hits the array of photodetectors. However, it is necessary to be certain that internally reflected light will not impact another photodetector

with sufficient intensity to turn it “on” (See Figure 4.) The exact quantification of the angles, distances, and intensities involves two basic equations. The first, is Snell’s law n 1 sin  1 = n 2 sin  2 (1) where n is the index of refraction and  is the angle from the perpendicular. The second is the Fresnel reflection coefficient that determines the fraction of intensity reflected at an interface [1]. cos cos[sin ( sin )] cos cos[sin ( sin )] 1 2 1 1 2 1 2 1 2 1 1 2 1 2     n n n n n n n n    

The second phase, the photodetector design and layout, is the heart of the spectrometer. Utilizing a unique geometry, the array of photodetectors (See Figure 2) is sensitive to the continuous spectrum, a feature not possible with other solid state layouts or discrete devices mounted on a circuit board. Another aspect to the design was to determine the currents created by each p-i-n. The optical generation current, I opto, is determined by Iopto  qAG 0 W (3) where A is the device area, W is the width of the intrinsic region, and G 0 is the steady-state generation rate of carriers per unit volume [2]. G 0 can be calculated by unit conversion for a known intensity I 0 , quantum efficiency , minority carrier diffusion length L p, and wavelength  hc G I Lp  0 ^0  (4) This optical generation current is created in addition to the reverse saturation current I off associated with all devices involving a reverse biased pn junction. The I off for a p-i-n is given by Sze [3] as p p off n L D I  qAp (^0) (5) The third phase, the fabrication of the array, entailed designing the process, creating the mask set and then fabricating. Fabrication was conducted in the ECE 344 Fabrication Facility at the University of Illinois. The process design rules emphasize speed and robustness to keep cost down and the yield up. First, all furnace steps should be as short as possible, but over 15 minutes to keep transients to a minimum. Additionally, short drives keep the sheet resistances R s and the resistivity  low. This reduces

2. OPTICS SUBPROJECT

2.1 Design Procedure (JSK) The optics portion of the design entailed separating the incident light source into its spectrum for analysis by the photodetector array. A prism was chosen because of its straightforward refractive properties that were ideal for this purpose. Snell’s Law, Eq. (1), allows the calculation of where certain wavelengths will hit the photodetector array as shown in Fig. 1. The next task was to determine where and at what angle the incident light would strike the prism. Shining the light on the center point of the face at an angle above the perpendicular was chosen as in Fig. 1. This translates into large transmission coefficients and thus a greater portion of the incident light hitting the photodiodes. Once this was set, Snell’s Law and basic geometry were used to determine where each wavelength would hit the silicon wafer for different angles of incidence. This was done on Mathematica. The extrema wavelengths 400nm and 700nm must hit the edges of their respective devices as shown in Fig. 1. This requires the spread to be an integer number of device widths wide. 2.2 Design Details (JSK) From the Mathematica code results, all the angles of the geometry (See Figure 1) are known as are the wavelength ranges corresponding to each device. These results are summarized in Table 1. TABLE 1. WAVELENGTH RANGES OF PHOTODETECTORS Photodetector Number Starting Wavelength (nm) Ending Wavelength (nm) Range (nm) 5 400 419 20 4 420 459 40 3 460 506 47 2 507 567 61 1 568 700 133 The nonlinearity of the optics is evident from the variation of the range over the spectrum. Also found from Mathematica is that the first photodetector is 0.0261m in from the prism edge. Additionally, using the data in Table 1, the average resolution of the photodetectors to be 60.2nm. This is only 0.2nm short of the 60nm specification given in the Proposal despite having to double the minimum feature size.

Using these angles and the modified Fresnel reflection coefficient given by Eq. (2), it was possible to analyze some potential sources of noise. Feedback in the prism is illustrated by Fig. 4. A portion of the light heading for the photodetectors is reflected off the prism faces and hits the wafer at a different position than the initial transmitted light. The cavity feedback is where the light reflected off the silicon surface bounces back off the prism face and again impacts the silicon (See Figure 4.) Since  is ~0.1 over the range of incident angles, each interface attenuates the feedback intensity, and hence the feedback current, by roughly an order of magnitude. Hence the noise voltage created across R 1 in Fig. 10 is at most an order of magnitude less than V on. This is easily filtered out by the comparator circuit.

electron hole pairs (EHPs) do not recombine in the intrinsic region. This criterion is shown in Appendix A. The size of the devices is governed by the minimum feature size. To minimize cost, the masks used in fabrication were printed on a 5080dpi laser printer. This fixed the minimum feature size and the device size. Using this dimension, the number of devices and the aforementioned angle of incidence were found such that the maximum number of devices fit exactly in the spread of light on the wafer. With the layout known, the expected currents can be calculated. The optical generation current produced by an incident light source of intensity I 0 can be found by substituting Eq. (4) into Eq. (5). This method assumes that the silicon absorbs the full intensity. Examining the light incident on the silicon involves reflection and transmission at two interfaces. Using Eq. (2) twice, reflection coefficients for both interfaces Air/SiO2 and SiO2/Si can be calculated. The intensity of light creating carriers I ’ 0 is I 0 scaled by the transmission coefficients. In other words, I ’ 0 is / 2 2 / 0 / 2 2 / 0 ' I (^) 0 ( 1  AirSiO )( 1   SiO Si ) I   AirSiO  SiO SiI (9) Another scaling factor needed to modify the I opto is due to the minority carrier recombination in the i-region. Thus, the optical generation current generated by an incident light source on the photodetectors is given by Eq. (4) and Eq. (5) with I 0 replaced by I ’ 0 and scaled by the 0. recombination factor. This gives a theoretical value of 0.7809A for I opto. The other current needed is the reverse saturation current Ioff. This current is present when the device is illuminated and when it is dark. I off is found to be 0.676A by straightforward application of Eq. (3), Eq. (4), and Eq. (5). From these current values, the current expected when the p-i-n is exposed to intensity I 0 is I (^) on  Ioff  I opto (10) Comparing these predicted currents to measured currents will give an indication of p-i-n performance.

4. FABRICATION OF PHOTODETECTORS

4.1 Design Procedure (SRB) The design of the process was driven by the desire to achieve nearly ideal devices, while keeping cost at a minimum. Additionally, process robustness was emphasized to lower production cost by ensuring a high yield. Common processing techniques and rules outlined in the ECE344 Lab Manual [4] were followed exactly. 4.2 Design Details (SRB) The first step is the removal of native oxide. An etch in the buffered oxide etch (BOE) solution was substituted for the more lengthy etch and RCA clean procedure to save time. The second step was growth of field oxide. A thick oxide is desired because it provides a good mask and to reduce effects of overetching. To obtain a thick quality oxide, the ECE 344 field oxidation sequence was used. It consisted of a long steam oxidation between short dry oxidations [4]. In order to reduce the driving of dopants, the phosphorous predeposition (predep) was performed first. The furnace temperatures in the ECE 344 facility are fixed and the phosphorous predep is at a higher temperature than the boron predep. Doping in phosphorous first leaves the lower temperature predep second. The phosphorous will be driven less than the boron would if it was doped in first due to the temperature difference of the second predep. As shown in Appendix A, depthwise fringing of the electric field is negligible. Thus, drives are not needed to push dopants into the wafer because little is gained by increasing the depths a few micrometers, when compared to W. The first drive was needed to regrow oxide to act as a mask for the next photoresist (PR) step. A typical sequence for patterning contacts is that following the final predep, oxide is regrown with a drive, then a PR step is used to open windows in the oxide. Another PR step is performed and aluminum is evaporated on top. The unexposed PR and the overlying aluminum are lifted off with acetone leaving the aluminum contacts. For the p-i-n process, a different sequence was developed [4]. The drive and first PR are replaced by a simple flood etch to remove all oxide. Then PR

5. EXTERNAL CIRCUIT

5.1 Design Procedure (SRB) The design of the external circuit is governed by its desired properties. When incident light falls on a photodiode, the circuit should light the appropriate LED. However, the circuit must be capable of discriminating against a false positive caused by noise sources discussed earlier in Section 2.1. From this, it became clear that simply amplifying a voltage created by the diode current and using it to drive the LEDs would be insufficient. Some type of filtration was needed. To this end, a comparator was used. The voltage created by the diode current dropped across a resistor is measured against a threshold value. If the voltage is above threshold then the LED is turned on; otherwise it is off. This offers the capability of filtering out any false positives simply by raising the threshold value above the noise floor. For the purposes of constructing the prototype, potentiometers were the ideal choice for generating the threshold voltage and R 1. The flexibility was required because the actual photodiode current values, produced by the fabricated devices, were unknown during design. It is important to note that voltages across R 1 when the illumination is on and off, V on and V off, must be small in order to justify the assumption that all 5V falls across the actual photodetectors. This requirement sets a ceiling for the maximum value of the potentiometer used to create these values. As stated in the Introduction, Eq. (6) is used to set the threshold voltage. From Section 2.2, the noise floor is only a few percent of the optical voltage above V off. Since Eq. (6) sets the threshold at 25% of the optical voltage above V off, no noise source is capable of creating a false positive output. Using these guidelines, Fig. 10 was designed. The hystersis resistor, R 2 , is required to pull the output up to a high voltage level [5]. The resistor R 3 , is required to prevent the current from becoming too high and destroying the LED. Five copies of the circuit are needed, one for each p-i-n. This design provides a simple, yet highly reliable, method for displaying the spectrometer results. 5.2 Design Details (SRB)

The first step in specifying component values for the external circuit is to determine R 1. One constraint on the resistor value is the voltage across it must always be small relative to 5V. This ensures the accuracy of the assumption of a 5V reverse bias across the photodetector when determining I 0 and I opto. The other constraint on the value of R 1 is that the voltages created must lie in the range where the comparator is sensitive. Because of its strong sensitivity, the MC3302 was chosen as the comparator [5]. This device allowed the voltages to be quite small. The resistance R 1 was chosen to be 10. From Ohm's Law and the predicted current values found in Section 3.2, the expected values of V on and V off were found to be 14.65V and 6.76V respectively. Using Eq. (6), V th was determined to be V V V V V V V on off off on th off^8.^71 4 3 4       (11) To create this voltage, a potentiometer with range 0-1M was used. To create R 1 , a 0-500 potentiometer was used. The last aspect of the external circuit design is to determine the values of the other two resistors in the circuit. The hystersis resistor R 2 was chosen to be 10k as recommended by Figure 2 in the MC3302 data sheet [5]. The other resistor R 3 was chosen to be 1k from past experience with similarly configured LED circuits.

7. DESIGN VERIFICATION

7.1 Sheet Resistance Measurements during Fabrication (JSK) After the borosilicate and phosphosilicate glass etches, R s was measured and compared to the ideal values calculated by Difcad (See Figure 6.) Table 2 compares the simulated R s values with those measured on the 4-point probe. TABLE 2. SHEET RESISTANCE MEASUREMENTS COMPARED WITH DIFCAD After Removal of R s Simulated (/ ) R s Measured (/ ) Phosphosilicate glass 30.292 30. Borosilicate glass 113.081 105. Both are sets are in near perfect agreement. 7.2 I opto and I off Measurements (SRB) After fabrication, I opto and I off were measured, on the test stations in the ECE 344 Lab and their values compared to the values predicted by the theory. This involved two dc sweeps to determine the illuminated and dark I-V characteristics of one p-i-n. The first was a sweep from –10V to +10V of the device in total darkness. This gave I off plotted versus V (See Figure 11.) Figure 12 was given by an identical sweep with the photodetector illuminated by a 628nm 1mW laser. The difference between the two curves is I opto. The spectrometer was designed to operate with a power supply of 5V. Table 3 shows that the values at this bias agree quite favorably with those predicted. TABLE 3. I off MEASUREMENTS COMPARED TO PREDICTED VALUES Current Predicted (A) Measured (A) I off 0.676 0. I opto 0.7809 0. The variations between measured and predicted values are easily reconciled by considering the finite resistance in the p and n wells along with the inherent inaccuracies along with the slight inaccuracies in the processing.

What remains is to be certain that the fabricated devices are within the tolerances required by the spectrometer. From Section 5.2, the threshold voltage, determined from the theoretical current values is 8.71V. From Table 3, the tested values of V on is 14.65V and V off is 6.76V. Since V th lies between V in and V off, the photodetectors are within tolerances and are suitable for use in the spectrometer. 7.3 Test of External Circuit (JSK) The final aspect of testing involved testing the external circuit to show it to be capable of use in the spectrometer. Initially the 0.40cm x 0.40cm pads were intended to allow easy access to each individual device. However, a poor aluminum liftoff created breaks and shorts between the pads. This is quite obvious in the photograph of Fig. 9. This required that probes be dropped directly over the actual wells. Because of the ECE 344 Lab computer problems, it was impossible to use only a single device with the external circuit because the test stations were needed for such accurate probing. To compensate, the n-wells were shorted together using conductive epoxy, as were the p-wells, forming one large p-i-n photodetector. Wires were then bonded to the large device. With R 1 = 11.1 (See Figure 10), the “on” voltage was found to be 6.61mV while the dark voltage was 4.60mV. From this V th was determined to be V V V V V V V on off off on th off^5.^12 4 3 4       (13) Tuning the threshold potentiometer, V th was set to 5.13V. The LED was lit when the p-i-n’s were illuminated by the 1mW 628nm laser and dark when the photodetectors were dark. This was demonstrated to Professor Swenson and Ben Floering on 4/27/99. Thus it is clear that aside from the poor aluminum liftoff, the photodetector array functioned properly as did the external circuit.

9. CONCLUSIONS (JSK)

While there were several setbacks involved with building the spectrometer, the design is sound. The liftoff problems can be solved by using a negative PR with a lightfield mask. With this change, the design will work and have 60.2nm of average resolution. However to make it of commercial quality, several changes must be made. To increase resolution to the range needed, a transmission grating must be used for dispersion rather than a prism. Further increases in resolution may be attained by using a commercial quality mask set with a much smaller feature size. This will also allow the widths of the photodetectors to be varied to compensate for the nonlinearity of the optics, giving the devices a uniform wavelength range. Additionally, a calibration feature is needed. This is easily performed by finding the largest I illuminated and using it with the theoretical value of I off to set V th. Calibration would allow the spectrometer to operate using light sources of varying intensities to illuminate the substance under test. Additionally, collimating optics should be added on the front end of the design because the design presented in this report assumed laser sources. Finally, to reduce mass-production cost, the external circuitry can be integrated onto the same wafer as the p-i-n array. These modifications will make the spectrometer on a chip a competitive product.

FIGURES

Fig. 1. Spectrometer overview. Fig. 2. Top view of photodetector array. Fig. 3. Block diagram of spectrometer design. Fig. 4. Feedback of light. Fig. 5. Finger layout. Fig. 6. Difcad simulation. 60 o 60 o 1 2 3 4 5 p-i-n photodetectors   air prism silicon Incident light  1  2 n n n n n n n n n p p p p p p p p p Vcc i-region  (^) lines 40 m 40 m 60 o 60 o 1 2 3 4 5 p-i-n photodetectors   Incident light

i

n

p

i

n

p

i

n

p

Vcc

Vcc

Vcc

DIFCAD 2 (ver 2.1) n-wells

Process Summary Species Nsurf Time (sec) Temp (C) BackgroundBoron (B) 1.90E+ Predep #1 Phosphorous (P)1.00E+21 900 1000 Drive #1 1219.8 1100 Predep # Drive # Predep #1 Characteristics P SOA Exact Xj 0.358 0.358 (um)  (^) 921.597 921.598 (-cm)^- Rs 30.292 30.292 /square Drive #1 Characteristics SOA Exact N^01 1.57E+20 1.57E+20 (cm^-3) Xj 1.456 1.456 (um)  (^) 417.195 417.197 (-cm)^- Rs 16.458 16.458 /square Optics Final Product Fabrication of Photodetectors External Circuit Design/Layout of Photodetectors