A precision integral resistor process has been successfully developed using a 10 sz/sq. tantalum nitride thin film. Although the integral resistors are overcoated.
Precision Embedded Thin Film Resistors for Multichip Modules (MCM-D) Chang-Ming Lin, Elizabeth A. Logan, and David Tuckerman nCHIP, Inc. 1971 North Capitol Avenue San Jose, CA 95132 nominal sheet resistivity and temperature coefficient of resistance (TCR) of this tantalum nitride film are 10msquare and -60 f 30 ppm/"C, respectively. The film is fabricated via a reactive DC magnetron sputtering process and subsequently defined by photolithography to form the many embedded resistors. The layer on which these are formed is beneath the signal layers by design and so is buried under six microns of silicon dioxide. With this overlying oxide, the buried resistors are well protected from any corrosion and mechanical damage. A test vehicle with various types of resistor geometries was designed and fabricated for evaluating the laser trimming process. The distribution of as-fabricated resistance of these resistors was found to be typically around 5 10% over a five-inch wafer. To further tighten this resistance distribution, a precision resistor trimming method is needed. This trimming process must be able to increase the resistance without damaging the MCM-D substrate (specifically, the surrounding SiOz and the integral capacitor which lies below the resistor). Laser trimming appeared to be the ideal candidate. Because of the 6 pm of overlying oxide, the conventional laser ablation method is not feasible. An advanced laser trim was thus contemplated.
Abstract A precision integral resistor process has been successfully developed using a 10 sz/sq. tantalum nitride thin film. Although the integral resistors are overcoated by 6 pm of PECVD silicon dioxide, a precision laser trimming process was developed which is capable of trimming the embedded resistors to 50 L2 with an accuracy of better than 50.5 L2 (1%) and with no damage to the surrounding structure. The stability of the trimmed resistors has been demonstrated and the average post-trim TCR value can be improved by up to 33%, depending upon the characteristics of the laser system. Trimmed integral resistors have also been examined by transmission electron microscope (TEM). Secondary grain growth within the trimmed resistor and spherical inclusions in the oxide near to trimmed resistor regions were observed by this analysis. As part of a reliability evaluation, the trimmed resistors were subjected to a severe manual thermal shock test over a AT of -500°C without catastrophic failure.
Introduction As the applications for MCM-D technology enter the regions of high speed and sensitive analog circuitry, the requirement for precision resistors becomes critical and essential. The use of discrete resistor chips cannot provide optimum impedance control at high frequencies, due to capacitive and inductive parasitics. Furthermore, the use of such devices not only consumes substrate area, increasing size and cost of the module, but also mean that the assembly of an MCM becomes more time consuming and challenging. All of these problems can be solved or minimized simply by integrating these resistors into the MCM-D substrate. This makes the integration of precision resistors an important extension to existing MCM-D technology. A tantalum nitride thin film resistor layer has been proved to be compatible with the aluminum and silicon dioxide process of nCHIP MCM-D substrates [I]. The
Sample design and process description A test vehicle was specially designed with various shapes, widths, lengths, values and inter-resistor pitch of resistor designs in order to evaluate the laser trim process. The nominal resistance value of these resistor designs was intentionally designed to be 10% lower than the desired value, with the interest of trimming "up" to value. This process starts with oxide growth and the construction of the integral decoupling capacitor, which is subsequently coated with 7.5 pm of PECVD SiOz. On top of these layers, a 1OL2/sq. tantalum nitride thin film is deposited, together with a metal signal layer, via a reactive DC magnetron sputtering process. The uniformity of sheet resistivity of this as-deposited film is better than -t 5% over a five inch wafer, with an average
44 0-8186-7789-9/97 $10.00 0 1997 IEEE
resistor stability, and damage to the PECVD SiOz and integral decoupling capacitor. Visual inspection was conducted under an optical microscope at a magnification of at least 50X. Trimmed samples were also studied by SEM and TEM after a brief cold “Vapox etch (a solution mixture of ammonium fluoride, acetic acid, and DI water). It is estimated that about 0.2 ym oxide would be removed from the Si02 surface by this etching decoration, and so this etch would reveal any hidden micro-fissures. The post-trim resistance was examined by four-point manual probing. A set of optimal trimming parameters was therefore established for both laser systems. With these optimal trimming factors, test samples were produced for thermal shock test and constant-current stress testing. Differences in the trimming resolution and capability of the laser systems were observed. Resistors trimmed by laser A had a better accuracy of post-trim resistance than laser B (Table 2), probably due to the elimination of contact resistance by the four point probing method. Laser A also gave better resistor stability (Figure l), and less impact on the surrounding SiOz (Figure 2). Typically, the accuracy was better than 1% for a trimmed 50 R resistor. SEM micrographs showed no sign of oxide damage. A TEM examination (Figure 3) showed many spheres included in the PECVD oxide near the interface with a trimmed resistor region. The result of elemental analysis showed the main constituent of these spherical inclusions to be tantalum (Figure 4). This phenomenon is consistent with the theory proposed by Muller and Mickanin [4]. Metal vapor and gas released from the thermally decomposed tantalum could deform the oxide envelope (Figure 2) and the heat content of the metal vapor may melt the surrounding PECVD oxide for a brief moment, allowing tantalum vapor to migrate into the Si02 layer and quench into spheres. At the same time, the dissociated nitrogen and/or oxygen could expand within the melted tantalum nitride, giving sporadic bubbles which froze inside the tantalum nitride layer when temperature was lowered. No material is lost during this laser trimming, and the resistance change is not solely due to cutting through the resistor layer. A similar phenomenon was observed in resistors trimmed by laser system B.
TCR of -60 ppm/”C which is consistent with previously published data [2,3]. This resistive layer is subsequently defined by photolithographic techniques. Above the patterned resistor and metal signal layer, a further interlayer dielectric, metal signal layer, and passivation layer are deposited and patterned. After fabrication, an automatic resistance test was applied to screen the integral resistors via a two-point probing method. Test results showed an excellent resistance accuracy for resistors with straight-bar design (Table 1) and the total decoupling capacitor yield was consistent with designs not incorporating a resistor layer. A much bigger deviation in resistance compared to the design value was observed for complex resistor designs, such as serpentine structures. This was due to the difficulty in modeling the resistance value for shapes other than a straight bar. The smallest space between resistors on the test vehicle is 10 pm and, as no difficulties in processing were encountered, it is likely that the minimum achievable inter-resistor space is less than 10 pm.
*
Laser trimming trials Conventional chip resistors are generally unpassivated and trimming is conducted by removing resistor material via laser ablation. This method becomes inappropriate for embedded resistors with 6 pm of overlying PECVD oxide which needs to remain intact. A study on trimming passivated resistors was reported by Muller and Mickanin [4] who trimmed NiCr film overcoated with -1 pm PECVD nitride. In those trials, the metal vaporized, deforming the PECVD nitride film and disrupting the NiCr film. Passivation cracking and delamination problems were also reported. For this reason, trimming the embedded resistors using a phase change or thermal decomposition approach was contemplated so that, with an appropriate laser power density, the damage to the PECVD oxide might be eliminated. To do this would require accurate control of the laser conditions, e.g., laser power. A trim trial matrix was generated and applied to the aforementioned test vehicle using two different laser systems. Both laser A and B operate at the fundamental wavelength of 1064 nm from a Nd-YAG crystal. A diode pumped laser source was used in laser A, and a continuous wave (CW) in system B. Active resistance trimming was performed with a four-point probecard for system A and a two-point flying-head prober for system
(Note: *System A is a TLSI M310 laser and system B is TLSI M614.)
Results of characterization tests Various characterization tests were performed on the trimmed resistors. These include TCR measurement, post-trim resistance stability, TDR measurement, current stress testing, and a manual thermal shock test. To compare the resistance stability of both trimmed and untrimmed integral resistors, a high temperature anneal test was carried out on both trimmed and untrimmed
B. A feasibility study was conducted on the test vehicle, and the trimmed resistors were evaluated by optical
microscopy, resistance testing, and SEM and TEM analyses according to criteria such as resistance increase,
45
erratic, with the resistance increasing at first and then drifting back to around the initial value (Figure 1). After a 450°C anneal, the reaction of the untrimmed resistors to prolonged current stressing remained unchanged. A similar effect was also observed for integral resistors trimmed by laser A (Figure 7). In this case, the integral resistors trimmed by laser B behaved similarly to the untrimmed resistors, i.e., the resistance decreased slightly in the beginning and then became stable.
samples. The annealing was conducted with nitrogen at 450°C for an hour in a diffusion anneal furnace_. The results are given in the following.
Post-trim resistance stability Both short-term and long-term post-trim resistance stability was studied for resistors trimmed by both laser systems. The post-trim resistance of resistors trimmed by laser A appeared to be more stable and accurate than those of laser B. The resistance of samples trimmed by laser A could be increased by 1 to 2 Q after annealing (Figure 5), but this was not true for samples trimmed by laser B. Occasionally, a reduction of resistance was observed for samples trimmed by laser B (Figure 6).
TDR measurements TDR measurements were taken from several nominally identical integral resistors trimmed to 50Q by laser A. These samples were trimmed with different trimming patterns, such as multiple 100% plunge cuts across the width of resistors, multiple shaving cuts along the current direction of the resistors, L-cuts, and serpentine cuts. The resistance measured by TDR was consistently greater than that measured by manual four-point probing by about 3 Q (Table 4). This difference is probably due to the contact resistance associated with the probes used in the TDR measurement. No apparent effect of trimming pattern on the TDR resistance of these integral resistors can be concluded at this time.
TCR measurements The average pre-anneal and post-anneal TCR values of trimmed and untrimmed integral resistors were evaluated over temperatures from room temperature to 125°C (see Table 3). No significant change between pre-anneal and post-anneal TCR was found for untrimmed resistors (61.98 f 3.70 ppm/"C and -65.07 f 1.66 ppm/"C respectively) which demonstrates the thermal stability of these integral resistors. This property is also consistent with previously published data [2,3]. The pre-anneal and post-anneal TCR property was reduced by about 30% for the integral resistors trimmed by laser A, (-41.53 f 4.81 ppm/"C and -41.53 f 4.81 ppm/"C respectively). This can be attributed to the decomposition of the tantalum nitride by the laser energy and the growth of tantalum grains within the trimmed regions. The average of preanneal and post-anneal TCR value of integral resistors trimmed by laser B was similar to those of the untrimmed integral resistors, (-67.48 k 1.98 ppm/"C and -63.04 k 12.29 ppm/"C respectively). This may imply some undertrimming of the samples.
Thermal shock testing Thermal shock testing was carried out on trimmed resistors at the wafer level by manually cycling between liquid nitrogen (-196 "C) and a 300°C hot plate for 100 cycles. The dwell time at the hot and cold temperatures was long enough for the wafer to reach thermal equilibrium and the transfer time was controlled to less than 2 seconds. The AT of this thermal shock test was about 500°C and should have screened out any weakened interface between the tantalum nitride and silicon dioxide induced by the laser trimming. The resistance of these samples was taken before and after this thermal shock test. The results showed no change in the resistance of samples trimmed by laser A (Table 5) and no cracked PECVD SiOz. There was only one incidence of failure from samples trimmed by laser B. This may be because the width of the failed integral resistor, 10 pm by design, was much narrower than the spot size of the laser beam. This may have caused the resistor to be severely damaged during trimming, causing it to open as a result of thermal shock .
Stress testing with a constant high current density Since this laser trimming process utilizes a thermal decomposition mechanism to change the resistance of the tantalum nitride thin film, the stability of trimmed resistors may be affected. To establish if this was indeed the case, the trimmed and untrimmed samples were subjected to prolonged, constant high-current density stressing, using a current of 2 A/mm of resistor width. For the pre-annealed sample group, the resistance of the untrimmed resistor showed almost no change during this prolonged current stressing. For integral resistors trimmed by laser A, the resistance decreased by about 0.1 Cl initially and then leveled off at a stable value. The behavior of integral resistors trimmed by laser B was
Conclusions Successful integration of tantalum nitride thin-film resistors into nCHIP's MCM-D substrates has been demonstrated. A precision laser trimming process has
46
been established for such integral resistors embedded under 6 pm of PECVD S O z , which allows laser trimming without oxide damage. A laser system has been identified which can achieve an accuracy of k 1 % for a targeted 5042 trim. Post-trim resistance drift and stability have been proven to be minimal and the performance of post-trim resistors shows no degradation during prolonged constant high-current stressing. A calculation, based on the results obtained in this study, showed that a 25 pm square of this integral resistor can handle steady power up to 25 mW with no stability problems.
[l] Tuckerman, D., D. Benson, H. Moore, J. Horner, and J. Gibbons, ”A High-Performance Second-Generation SPARC MCM”, Proc. Intcrnational Conference and Exhibition on MultichiD Modules (ICEMCM), April 1994, pp. 314-319, Denver.
[2] Muller, M. 3. and W. Mickanin, “Functional Laser Trimming of Thin Film Resistors on Silicon ICs, sPIE Vol. 5 1 1 Laser Processing of Semiconductors and Hybrids, 1986, pp. 70-84. ”
[3] Berry, R. W., P. M. Hall and M. T. Harris, Thin Film Technology, Van Nostrand Reinhold publisher ,New York, 1968.
Acknowledgment The authors would like to acknowledge the support of DARPA for this work and the Sandia National Laboratories for their contribution to thc laser trimming trials and TEM analysis.
[4] Au, C. L., W. A. Anderson, D. A. Schmitz, J. C. Flassayer, and F. M. Collins, “Stability of Tantalum Nitride Thin Film Resistors,” J. Mater. Res., Vol. 5, No. 6, June 1990, pp. 1224-32.
References Laser A Laser B
Table 1. Comparison of designed resistance with asprocessed values.
Note: Units in ppm/”C.
Table 4. Comparison of resistance measured by TDR and four point probing.
Design Width Lot A Lot B Lot c ( W (Q) 40 9.99 f 0.09 9.75 ? 0.12 9.98 f 0.15 10 50 45.37 f 0.33 44.18 f 0.44 45.96 f 1.23 45 40 99.60 f 0.82 96.77 f 1.04 101.28 f 3.3 100 2000* 10 1267 f 29 1221 f 37 1230 rf: 47.3
Table 2. Comparison of resistance before and after trimming by laser A.
Note: Resistors were trimmed by laser A.
Table 5. Resistance data before and after thermal shock test
Pre-trim
t
45.31 & 0.53 Post-trim* I 50.05 +_ 0.07 Post-trim** 47.42 f 0.66 Note: 1. Data measured by 4-point probing. 2. * laser A system. 3. ** laser B system.
I - 41.53 rf: 4.81 I - 47.15 f 4.30 I - 67.48 f 1.98 I - 63.04 -C 12.29
I
Table 3. Average pre-anneal and post-anneal TCRs from trimmed and untrimmed resistors.
Note: Resistance taken by four-point probing method and samples trimmed by laser A.
Sample I Pre-anneal I Post-anneal Untrimmed I - 61.98 rf: 3.70 I - 65.07 f 1.66
47
1
I
I
I
I
I
I
I
I
b
.
I
I
O
I
8-
765 -
1-
-1
* *.*
3-
1 I
uY... . 8
210
2 ,
0
0
2
4
6
8
IO
Fig. 1. A comparison of resistance stability of trimmed and untrimmed resistors under high current density stressing
i
Fig. 5. Resistance change due to high temperature anneal, resistors trimmed by laser A.
Figure 2. TEM micrograph of post-trim PECVD §ioz cross section.
i
J
Figure 6. Resistance change due to high temperature anneal, resistors trimmed by laser B. Figure 3. Spherical inclusions in SiOz near the interface with a trimmed resistor region (TEM micrograph).
48
...
U"
_I__
I .
"(h.,
Figure 7. A comparison of resistance stability of trimmed and untrimmed resistors after anneal, under high current density stressing.
49
