Advanced electrical and stability characterization of untrimmed and ...
Microelectronics Reliability 46 (2006) 352–359 www.elsevier.com/locate/microrel
Advanced electrical and stability characterization of untrimmed and variously trimmed thick-film and LTCC resistors Andrzej Dziedzic a
a,*
, Andrzej Kolek b, Waleed Ehrhardt c, Heiko Thust
c
Faculty of Microsystem Electronics and Photonics, Wroclaw University of Technology, Wybrzeze Wyspianskiego 27, 50-370 Wroclaw, Poland b Department of Electronics Fundamentals, Rzeszow University of Technology, W. Pola 2, 35-959 Rzeszow, Poland c Microperipheric Group, Ilmenau Technical University, D-98684 Ilmenau, Germany Received 11 October 2004; received in revised form 30 November 2004 Available online 21 January 2005
Abstract As-fired thick-film resistors have the resistance tolerance within ±20% and this tolerance is increased for smaller components. Therefore the novel trimming methods are necessary for microresistors, especially when they are embedded in LTCC substrate. This paper compares electrical (normalized temperature dependence of resistance, low frequency noise) and stability properties (relative resistance drift, changes of current noise index) of untrimmed, voltage pulse trimmed and laser trimmed unglazed thick-film resistors after step-increased long-term thermal ageing at 162 C, 207 C and 253 C. Moreover the effect of long term exposure (1000 h, 125 C) and thermal shocks (1000 shocks between 55 C and 125 C) is analysed for untrimmed and voltage pulse trimmed buried LTCC resistors. 2004 Elsevier Ltd. All rights reserved.
1. Introduction Currently laser trimming is the dominant trimming method of thick-film and LTCC (low temperature cofired ceramics) resistors. However, because of continuous trend in electronics industry for smaller, lighter, faster and cheaper products, a significant dimension reduction of thick-film and LTCC resistors as well as the fabrication of buried components is noted at present. From one side the laser trimming is insufficient for the smallest thick-film and LTCC resistors, from the other
*
Corresponding author. Tel./fax: +48 7135 54822. E-mail address: [email protected] Dziedzic).
(A.
side the tolerance of as-fired resistors is within ±20% and increases with reduction of dimensions. Since approximately 10 years an intensive search has been made for alternative trimming method. The most advanced research is connected with trimming by energy of high voltage pulses. In general, the susceptibility to high voltage pulses and electrostatic discharges is very important and has been investigated for thick-film resistors for almost 30 years [1–4]. Such investigations can be performed with the aid of single or series of ‘‘long’’ pulses (rectangular shape surges with 1–20 ms duration) [5,6] or by the series of ‘‘short’’ rectangular voltage pulses (with duration from several hundred nanoseconds up to several hundred microseconds). Tobita and Takasago [7] and then a group from the Ilmenau Technical University [8–11]
0026-2714/$ - see front matter 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.microrel.2004.12.014
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359
and a group from the Wroclaw University of Technology [12–15] tried to trim thick-film or LTCC resistors by pulse voltage method. Very recently the Ilmenau group has built and tested high voltage capacitor discharging generator allowing for controllable resistance changes of low and high ohmic thick-film resistors [16,17]. But still the quality of trimmed resistors is characterized only by drift of resistance, caused by various long- and short-term exposures. This paper presents electrical and stability characterization of untrimmed, pulse voltage trimmed or laser trimmed thick-film resistors by extreme temperature stress. It compares their electrical (normalized temperature dependence of resistance, low frequency noise) and stability properties (relative resistance drift, changes of current noise index). Variously trimmed resistors were subjected to the stepincreased long-term thermal ageing at 162 C, 207 C and 253 C. These investigations may give farther information for the optimization of HV-trimming process especially for high temperature applications. Moreover the use of such high stress temperatures could be useful in analysis of conduction mechanism in variously trimmed thick-film resistors – it is supposed that the applied voltage pulses lead to local changes of the network properties due to Joule heating and local capacitor discharges. But still there is no full theory, which will fully describe the nature of these changes. The first step in creation of such theory one can find in [18] where the trimmed resistor was treated as the regular matrix made of random resistors and capacitors. The various resistance changes resistance observed as a result of applied high voltage pulse can be understood by requiring two competing behaviours of the local elements: the decreasing and enhancement of the local resistances. The quantitative agreement between proposed model and experiment can be obtained both in the frame of standard as well as biased percolation attempt.
by about +20% above initial resistance. In this case a simple P-cut with a cut depth of about 15% of the resistor width was prepared. Second group of resistors were trimmed by energy of high voltage pulses in system shown in Fig. 1. The system consists on • discharge circuit of HV-capacitor creating very short pulses, • programmable digital multimeter for resistor measurement, • high voltage relay to switch the connection between either the HV-capacitor discharging circuit or the multimeter and the sample (resistor), • microcontroller for controlling the trimming process [17]. The results of thick-film resistor trimming using discharge circuit of HV capacitor point out the same effect on resistance as by previously used HV burst generator [please see eg. [11]]. This means that the low ohmic resistors increase (an effect of disconnection of some conductive paths due to the applied electrical energy) and high ohmic ones decrease their values during trimming (applying of HV energy to the high ohmic resistance causes local heating at conductive region and temperature increase even above 600 C; this leads to formation of additional conductive paths or regions and to resistance decrease). The 100 X/sq resistors were trimmed by changing the high voltage amplitude in the range from 200 to 700 V but keeping the pulse duration constant and equal to 0.8 ls; every increase of trimming voltage causes increase in electrical energy and the value of resistance increases as well. Larger electrical energy is necessary
Discharge Circuit of HVCapacitor
2. Test sample fabrication and trimming Test samples with five one-square resistive structures (0.5 · 0.5 or 1 · 1 mm2) have been used for trimming of thick-film resistors by energy of high voltage pulses and for most of electrical and stability measurements. A part of untrimmed and laser trimmed resistors have 0.25 · 0.25 mm2 planar dimensions. Commercial resistor pastes 100 X/sq and 10 kX/sq as well as Pd–Ag based thick-film conductors were screen printed on alumina through 325 mesh stainless screen, then dried and fired at 850 C in a standard firing profile. The resistance of as-fired thick-film structures was from the range 400– 500 X/sq for 100 X/sq resistor paste and 7–9 kX/sq for 10 kX/sq one. The Aurel NAVS-30 laser trimming system with Nd:YAG laser crystal was used for resistor trimming
353
Sheet resistor
Nanosecond Switch
microcontroller
Ohm -Meter
Trimming Data Base
HV-Relay
Fig. 1. Concept of high voltage discharge trimming and measuring system.
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359
for high ohmic resistors trimming. Therefore during trimming of 10 kX/sq resistors the 900 V voltage was kept constant but the duration of discharge time changed in the range from 5 to 70 ls. Increasing of pulse width means increase of electrical energy, which affects the resistance of the tested component. This means that the level of resistance changes after HV trimming was dependent on the strength of electric field in the case of low resistive components or the pulse duration in the case of high resistive ones. One should note, that in all of these investigations the resistors were unglazed, because we wanted to investigate the microstructures before and after trimming. LTCC resistors were buried in DP 951 Du Pont green tape. The 2 · 1, 1 · 1 or 1 · 2 mm2 resistors with Aubased terminations were made from three different resistor ink systems with sheet resistance from 10 X/sq to 10 kX/sq and fabricated according to tape manufacturer recommendation. Part of these resistors were subjected to high voltage pulse trimming (the trimming conditions were similar as for surface resistors). The trim target for these components was a resistance change of about 25%.
0
o
162 C
o
207 C
o
253 C
-4 ∆R/Ro [%]
354
-8 -12 -16 100 ohm/sq. -20
0
S2 - 1x1 mm2 S3 - 1x1 mm2 S5 - 1x1 mm2 S10 - 0.5x0.5mm2 100 200 300 2 S11 - 0.5x0.5mm S13 - 0.5x0.5mm2
400 500 t [h]
600
700
Fig. 2. Long-term stability of high voltage discharge trimmed 100 X/sq thick-film resistors (changes of resistance in trimming process: S2; +12.1%, S3; +6.0%, S5; +18.0%, S10; +3.1%, S11; +10.5%, S13; +19.0%).
high voltage pulse trimmed (HV) 162 °C
0
3. Long-term thermal ageing
207 °C
-1
253 °C
-2
∆R/ Ro [%]
Most often long-term thermal stability of thick-film resistors is characterized by the drift of resistance. Usually such a drift is characteristic for particular resistor systems, is dependent on ageing temperature and time and also on trimming method. But there are contradictory information about long-term thermal stability of high voltage trimmed resistors. Very good properties were reported in [7,8,11] but data presented in [10] have shown that 1 h keeping at 300 C created 3% relative resistance changes for pulse voltage trimmed samples and only about 0.1% for untrimmed ones. Investigations described in [14] have shown that long-term stability of HV exposed structures deteriorates with increase of number of pulses and is somewhat worse in comparison with laser trimmed or untrimmed ones. Similar conclusions were presented in [12]. Herewith we reported comparison of resistance drift induced by long-term thermal ageing at three different temperatures 162 C, 207 C and 253 C. The samples were kept at every temperature for about 200 hours. The relative resistance changes of high voltage discharge trimmed resistors are shown in Figs. 2 and 3. Long-term stability of untrimmed and laser-trimmed resistors is presented in Fig. 4. It is shown that untrimmed and laser-trimmed resistors are much more stable than voltage pulsed ones by extreme temperature stress. Untrimmed resistors exhibit resistance increase at all temperature levels. This increase is larger at the beginning of every temperature step but even after keeping the samples at 253 C it does
-3 10 kohm/sq. S2 - 1x1 mm2 S3 - 1x1 mm2 S5 - 1x1 mm2 S10 - 0.5x0.5 mm2 S12 - 0.5x0.5 mm2 S13 - 0.5x0.5 mm2
-4 -5 -6 0
100
200
300 400 t [h]
500
600
700
Fig. 3. Long-term stability of high voltage discharge trimmed 10 kX/sq thick-film resistors (changes of resistance in trimming process: S2; 13.0%, S3; 10.3%, S5; 5.4%, S10; 3.8%, S12; 10.5%, S13; 13.6%).
not exceed 1%. Very small laser trimmed resistors (0.25 · 0.25 mm2) have negative resistance drift after keeping at 207 C and 253 C but still within the range of ±1%. Resistors trimmed by energy of high voltage pulses show decidedly larger negative drift of resistance. This drift is much higher for unglazed 100 X/sq than 10 kX/sq structures and it depends on trimming level. Resistors with higher trim level are much less stable at every investigated temperature (but for applications up to 100 C the HV-pulse trimming method leads to practically acceptable results [8–11]). Moreover one should remember that this method is the only trimming
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359
untrimmed (UN) laser trimmed (L)
10 kohm/sq. 0.6
0.25x0.25 mm2 (UN) 0.25x0.25 mm2 (L) 1x1 mm 2 (UN) 1x1 mm 2 (L)
0.2 0.0
10 kohm /sq.; 1x1 mm2
1.02
R/R25
0.4
∆R/Ro [%]
1.03
o
untrimmed laser trimmed HV trimmed
1.01
1.00
162 C -0.2
207 oC
253 oC
0.99
-0.4
-150 -100
-50
(a)
-0.6
0
100
355
200
300
400
500
0 50 o T [ C]
100
150
600
t [h]
possibility for buried resistors. The explanation of this behaviour could be, that during the thermal ageing up to 253 C the conductive particles of high voltage trimmed resistors could crystallize and decrease their resistance value. More detailed structural investigations are necessary to explain completely such behaviour. This method is the only trimming possibility for buried resistors. For applications up to 125 C the HV-pulse trimming method leads to practically good acceptable results [8–11].
4. Normalized temperature dependence of resistance Resistance measurements of resistors in a wide temperature range are important for the analysis of conduction process. The temperature characteristics were measured automatically in the range from 170 C to 130 C. The Keithley 2000 multimeter interfaced to PC was used for data acquisition and presentation. The examples of normalized temperature dependence of resistance are shown in Fig. 5(a) whereas Fig. 5(b) presents differential temperature coefficient of resistance dR (TCRdiff ¼ RdT ) versus temperature. The results for 100 X/sq structures are qualitatively similar. In practice there are no differences between R(T) and TCR(T) characteristics for untrimmed and laser trimmed resistors. It should be noticed that HV-trimmed resistor has R(T) curve with minimum of resistance shifted significantly toward higher temperatures. The second difference appears above 100 C, where instead of expected resistance increase a noticeable drop of R is observed. The shift of a minimum on R(T) curve can be explained basing on the model proposed by Kozlowski [19]. He analysed the changes of the percolation path caused by appearing microcracks in thick-film resistors. The occurrence of the microcracks manifests itself as
TCR [ppm/K]
Fig. 4. Long-term stability of untrimmed and laser-trimmed 10 kX/sq thick-film resistors.
2
10 kohm/sq.; 1x1 mm
100
untrimmed laser trimmed HV trimmed
0
- 100 - 200 - 300 -150
-100
(b)
-50
0 o T [ C]
50
100
150
Fig. 5. Comparison of (a) normalized temperature dependence of resistance, (b) differential temperature coefficient of resistance for untrimmed, laser-trimmed and high voltage-trimmed 10 kX/sq thick-film resistors.
removing of certain bonds, randomly chosen from the initial critical percolation path. After such an operation a new percolation path is found and because of slightly different temperature dependence of a new ‘‘unit cells’’ in the percolation path the changes of the temperature dependence of resistance are induced. Kozlowski has shown both numerically and experimentally that the increase in density of microcracks caused a shift of the resistance minimum to the higher temperatures i.e. in the same manner as in Fig. 5(a). At this very moment there is no explanation for the curve bending at about 100 C. However this temperature may be important from the long-term stability point of view. Fig. 5(b) indicates the possibility of TCR shifting in high value resistors which can be useful and desirable in some applications.
5. Low-frequency noise Current noise index (CNI) is one of the most important parameters of resistors. But the information stored
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359
CNI ¼ 10ðlogðS ln 10Þ þ 12Þ
ð1Þ
An interesting observation is that the slope of logSV versus log f plots in Fig. 6(b) increases after the resis-
10 kohm/sq.; 1x1 mm2; HV(-4,71%); before ageing
10
13.73 V 11.16 V 8.59 V 5.15 V 2.59 V
-11
-12
2
S u [V /Hz]
10
10 10 10
-13
-14
-15
1
10
(a)
100 f [H z]
10 kohm /sq.; 1x1 mm2; HV (-4,71%); after ageing
10
2
in its value, sufficient for practical applications, is not sufficient to identify sources generating the fluctuations. For these purposes the measurements of power spectral density are necessary [20, and references therein]. The standard procedure of such low frequency noise spectroscopy is to examine the product fSV of frequency f and power spectral density SV of noise voltage. The common 1/f noise in fSV versus f plots appears as a flat background, whereas the features are associated with other kind of noises. Noise spectra of thick-film resistors usually possess a typical 1/f shape. Recently it was proven that also polymer thick-film resistors [21] as well as LTCC resistors both buried and surface [22] show Gaussian 1/f noise below 1 kHz. There are also papers [e.g. 23,24] where low-frequency noise measurements are used as a diagnostic tool for thick-film resistor evaluation. Even recently it was shown that low-frequency noise is more sensitive to the resistor degradation due to high-voltage pulse stressing than resistance itself [25]. However, this was shown for the first time and the number of experimental results is up to date not too large. In this paper we have measured and compared the low-frequency noise of untrimmed, laser trimmed and voltage trimmed 100 X/sq and 10 kX/sq thick-film resistors with 0.5 · 0.5 mm2 and 1 · 1 mm2 planar dimensions. After the measurements the resistors were thermally aged at 162 C for 170 h and then again subjected to noise measurements. Noise measurements were carried out by a standard DC technique. The sample of resistance R was biased through the ballast wire-wound resistor RB from a DC source. The noise signal was AC-coupled to the lownoise preamplifier Unipan 233-7 (bandwidth 0.5 Hz–100 kHz). Then it was sent to the HP 35660A digital signal analyser in which power spectral density SV of voltage dV was calculated in the frequency range from 2 Hz to 800 Hz for several voltages biasing the sample. After subtracting the noise background, SV=0 which was the sum of preamplifier noise and thermal noise, it has occurred that the spectra have 1/f shape (the line in log–log plot has a slope of approximately 1) (Fig. 6), Consequently fSV plots depend on frequency rather weakly (Fig. 7) and so, we use the average h fSViover the measured frequency range to estimate the frequency independent noise level. This quantity is still bias dependent. When logh fSVi is displayed versus logV (Fig. 8) the lines drawn through the data have the slope of approximately 2. This means that the relation SV V2 holds and the physical origin of the noise are equilibrium resistance fluctuations. The quantitative measure of such a noise is the ratio S = h fSVi/V2, which is related to CNI via the relation
Su [V /Hz]
356
10 10 10 10
11.89 V 8.93 V 5.96 V 2.92 V
-10
-11
-12
-13
-14
1 (b)
10
100 f [Hz]
Fig. 6. Power spectral density of ‘‘excess’’ low-frequency noise for high voltage trimmed 10 kX/sq thick-film resistor: (a) astrimmed (before ageing), (b) after thermal ageing (170 h, 162 C).
tors are thermally aged, and the plots in Fig. 7(b) are no longer flat. We leave the reasons for this behaviour for further studies. At this moment we summarize the effects of long-term thermal ageing on resistance drift and low frequency noise only on in a quantitative manner. This is done in Table 1. Data for low frequency are expressed in the terms of both power spectral density and current noise index. The changes in CNI are also shown in Fig. 9. It is interesting that the noise index of as-fired laser or HV-trimmed resistors is not worse than that of untrimmed resistors. After ageing we observe some insignificant variations of CNI for untrimmed or laser trimmed resistors. For HV-trimmed resistors, however, ageing process leads to noticeable increase of noise index. The increase of CNI (S) is larger in case of significantly trimmed structures (it was shown in Section 3, that HV-trimmed resistors with higher trim level are much less thermally stable). Let us note that while ageing the resistance of these (i.e. HV-trimmed) resistors has decreased by around 0.5–2%.
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359 2
2
10 kohm/sq.; 1x1 mm ; HV (-4,71%); before ageing
-10
2
f Su [V ]
10
-10
10
-11
B Linear Fit of Data3_B ###
2
13.73 V 11.16 V 8.59 V 5.15 V 2.59 V
f Su [V ]
10 kohm/sq.; 1x1 mm ; HV (-4,71%); before ageing
10
357
α =1.96
-11
10
10
-12
2
-12
10
1
10
4
6 U p [V]
(a)
100
8
10 12 14
f [Hz]
(a)
2
10 kohm/sq; 1x1 mm ; HV (-4,71%); after ageing
2
10 kohm/sq.; 1x1 mm ; HV (-4,71%); after ageing
2
f Su [V ]
-9
2
-9
10
10
f Su [V ]
11.89 V 8.93 V 5.96 V 2.92 V
-8
10
10
-10
10
-11
α =1.99
-10
10
10
-11
2
4
(b)
1 (b)
10
6 Up [V]
8
10 12 14
100 f [Hz]
Fig. 7. The product of frequency and power spectral density calculated for the data from Fig. 6(a) and (b).
6. Trim results and stability of buried LTCC resistors Long-term stability investigations were made after high-voltage-pulse trimming and after pre-conditioning. The latter was intended to simulate assembly processes.
Fig. 8. The dependence of averaged product of frequency and power spectral density on biasing voltage, fS V U aP (data for high voltage trimmed 10 kX/sq resistors) (a) before (b) after thermal ageing (170 h,162 C).
It contained 2· reflow soldering and next glob top curing for 5 h at 150 C. The final qualification tests are high temperature storage (1000 h, 125 C) and temperature cycling (1000 cycles from 55 C to 125 C).
Table 1 The effect of trimming method and ageing process on resistance drift and changes of relative power spectral density (S) and current noise index (CNI) Dimension (mm2)
Exposure Untrimmed Untrimmed Laser trimmed (+18.85%) High voltage pulse trimmed (1.9%) High voltage pulse trimmed (4.2%) High voltage pulse trimmed (4.7%)
Unaged Aged Unaged Aged Unaged Aged Unaged Aged Unaged Aged Unaged Aged
1·1 1·1 1·1 1·1 1·1 1·1
R (kX) 11.476 11.538 12.008 12.045 11.248 11.299 11.510 11.472 10.920 10.801 10.905 10.744
S(f=10 Hz)
CNI (dB) 14
2.88 · 10 4.32 · 1014 1.03 · 1013 8.07 · 1014 2.19 · 1014 1.76 · 1014 3.22 · 1014 6.03 · 1014 4.69 · 1014 2.05 · 1013 3.34 · 1014 5.50 · 1013
11.8 10.0 6.2 7.3 13.0 13.9 11.3 8.6 9.7 3.3 11.2 1.0
358
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359
.
.
.
.
frequency noise) and stability (relative resistance drift, changes of current noise index after ageing) properties of untrimmed, laser trimmed as well as high voltage trimmed thick film resistors. Moreover they were correlated with additional resistor high temperature ageing after trimming process. According to our knowledge this is the first so wide comparison of various trimming method. During these investigations the following was found:
Fig. 9. Changes of current noise index of 10 kX/sq as a function of trimming method and ageing process.
After pre-conditioning the first results of trim stability were obtained. Contrary to components on alumina trimmed buried LTCC resistors drifted always in the trimming direction i.e. they showed the positive drift if resistance increased after trimming increase and the negative drift if resistance decreased after this process. One candidate (A-100) changed values of more than 5%. Two other pastes, having mean deviations of about 1%, had one or two group members exceeding this value up to 6%. The control groups showed very low deviations. Table 2 shows that some of the used resistor systems are suitable for high voltage trimming. Some of them (C-10k, B-10k, C-100) have a deviation < 1% after preconditions and thermal ageing and others (A-10, A-100, A-1k) have a drift up to 6%. This observation points out that the trimming stability after thermal ageing is depended on resistor paste. 7. Conclusions In this paper we presented wide spectrum of electrical (normalized temperature dependence of resistance, low
• There is distinct difference in long-term thermal stability of investigated components. The drift of untrimmed and laser trimmed resistors is similar. But structures trimmed by series of voltage pulses after extreme temperature stress have stability worse by about one order of magnitude. The drift is larger for resistors more trimmed by pulses (having larger resistance difference before and after trimming). • The temperature dependences of resistance in a wide temperature range are similar for untrimmed and laser trimmed components. However the minimum of resistance in HV-trimmed resistors is shifted to the higher temperatures. Moreover unexpected change of R(T) curve shape appears near 100 C. • The noise with a typical 1/f shape caused by an equilibrium resistance fluctuations is dominant in the low frequency range independently of the trimming method. The initial (i.e. before ageing) noise intensity is comparable for all resistors made of the same ink and of identical planar dimensions. After long-term thermal ageing the current noise index of HV-trimmed resistors noticeably increases in spite of simultaneous slight decrease of resistance. The increase of CNI is larger in case of strongly trimmed structures. On the other hand the noise level of untrimmed and laser trimmed resistors does not undergo changes after ageing process (similarly as resistance). • Trimming of unglazed resistors by energy of high voltage pulses induces other ageing mechanisms than those, which are predominant in untrimmed or laser trimmed resistors, which is essential for high temperature load. Such trimming method is useful in many standard applications where the temperature of resistors do not exceed 100 C.
Table 2 Long-term stability of LTCC resistors versus applied resistive ink Paste
Rsq after three refirings and tolerance
Trim range and direction of resistance changes
A-10 A-100 C-100 A-1k B-10k C-10k
50 X/sq ±50% 40 X/sq ±50% 80 X/sq ±20% 130 X/sq ±20% 2.3 kX/sq ±25% 3 kX/sq ±30%
>25% >25% >25% >25%
Advanced electrical and stability characterization of untrimmed and variously trimmed thick-film and LTCC resistors Andrzej Dziedzic a
a,*
, Andrzej Kolek b, Waleed Ehrhardt c, Heiko Thust
c
Faculty of Microsystem Electronics and Photonics, Wroclaw University of Technology, Wybrzeze Wyspianskiego 27, 50-370 Wroclaw, Poland b Department of Electronics Fundamentals, Rzeszow University of Technology, W. Pola 2, 35-959 Rzeszow, Poland c Microperipheric Group, Ilmenau Technical University, D-98684 Ilmenau, Germany Received 11 October 2004; received in revised form 30 November 2004 Available online 21 January 2005
Abstract As-fired thick-film resistors have the resistance tolerance within ±20% and this tolerance is increased for smaller components. Therefore the novel trimming methods are necessary for microresistors, especially when they are embedded in LTCC substrate. This paper compares electrical (normalized temperature dependence of resistance, low frequency noise) and stability properties (relative resistance drift, changes of current noise index) of untrimmed, voltage pulse trimmed and laser trimmed unglazed thick-film resistors after step-increased long-term thermal ageing at 162 C, 207 C and 253 C. Moreover the effect of long term exposure (1000 h, 125 C) and thermal shocks (1000 shocks between 55 C and 125 C) is analysed for untrimmed and voltage pulse trimmed buried LTCC resistors. 2004 Elsevier Ltd. All rights reserved.
1. Introduction Currently laser trimming is the dominant trimming method of thick-film and LTCC (low temperature cofired ceramics) resistors. However, because of continuous trend in electronics industry for smaller, lighter, faster and cheaper products, a significant dimension reduction of thick-film and LTCC resistors as well as the fabrication of buried components is noted at present. From one side the laser trimming is insufficient for the smallest thick-film and LTCC resistors, from the other
*
Corresponding author. Tel./fax: +48 7135 54822. E-mail address: [email protected] Dziedzic).
(A.
side the tolerance of as-fired resistors is within ±20% and increases with reduction of dimensions. Since approximately 10 years an intensive search has been made for alternative trimming method. The most advanced research is connected with trimming by energy of high voltage pulses. In general, the susceptibility to high voltage pulses and electrostatic discharges is very important and has been investigated for thick-film resistors for almost 30 years [1–4]. Such investigations can be performed with the aid of single or series of ‘‘long’’ pulses (rectangular shape surges with 1–20 ms duration) [5,6] or by the series of ‘‘short’’ rectangular voltage pulses (with duration from several hundred nanoseconds up to several hundred microseconds). Tobita and Takasago [7] and then a group from the Ilmenau Technical University [8–11]
0026-2714/$ - see front matter 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.microrel.2004.12.014
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359
and a group from the Wroclaw University of Technology [12–15] tried to trim thick-film or LTCC resistors by pulse voltage method. Very recently the Ilmenau group has built and tested high voltage capacitor discharging generator allowing for controllable resistance changes of low and high ohmic thick-film resistors [16,17]. But still the quality of trimmed resistors is characterized only by drift of resistance, caused by various long- and short-term exposures. This paper presents electrical and stability characterization of untrimmed, pulse voltage trimmed or laser trimmed thick-film resistors by extreme temperature stress. It compares their electrical (normalized temperature dependence of resistance, low frequency noise) and stability properties (relative resistance drift, changes of current noise index). Variously trimmed resistors were subjected to the stepincreased long-term thermal ageing at 162 C, 207 C and 253 C. These investigations may give farther information for the optimization of HV-trimming process especially for high temperature applications. Moreover the use of such high stress temperatures could be useful in analysis of conduction mechanism in variously trimmed thick-film resistors – it is supposed that the applied voltage pulses lead to local changes of the network properties due to Joule heating and local capacitor discharges. But still there is no full theory, which will fully describe the nature of these changes. The first step in creation of such theory one can find in [18] where the trimmed resistor was treated as the regular matrix made of random resistors and capacitors. The various resistance changes resistance observed as a result of applied high voltage pulse can be understood by requiring two competing behaviours of the local elements: the decreasing and enhancement of the local resistances. The quantitative agreement between proposed model and experiment can be obtained both in the frame of standard as well as biased percolation attempt.
by about +20% above initial resistance. In this case a simple P-cut with a cut depth of about 15% of the resistor width was prepared. Second group of resistors were trimmed by energy of high voltage pulses in system shown in Fig. 1. The system consists on • discharge circuit of HV-capacitor creating very short pulses, • programmable digital multimeter for resistor measurement, • high voltage relay to switch the connection between either the HV-capacitor discharging circuit or the multimeter and the sample (resistor), • microcontroller for controlling the trimming process [17]. The results of thick-film resistor trimming using discharge circuit of HV capacitor point out the same effect on resistance as by previously used HV burst generator [please see eg. [11]]. This means that the low ohmic resistors increase (an effect of disconnection of some conductive paths due to the applied electrical energy) and high ohmic ones decrease their values during trimming (applying of HV energy to the high ohmic resistance causes local heating at conductive region and temperature increase even above 600 C; this leads to formation of additional conductive paths or regions and to resistance decrease). The 100 X/sq resistors were trimmed by changing the high voltage amplitude in the range from 200 to 700 V but keeping the pulse duration constant and equal to 0.8 ls; every increase of trimming voltage causes increase in electrical energy and the value of resistance increases as well. Larger electrical energy is necessary
Discharge Circuit of HVCapacitor
2. Test sample fabrication and trimming Test samples with five one-square resistive structures (0.5 · 0.5 or 1 · 1 mm2) have been used for trimming of thick-film resistors by energy of high voltage pulses and for most of electrical and stability measurements. A part of untrimmed and laser trimmed resistors have 0.25 · 0.25 mm2 planar dimensions. Commercial resistor pastes 100 X/sq and 10 kX/sq as well as Pd–Ag based thick-film conductors were screen printed on alumina through 325 mesh stainless screen, then dried and fired at 850 C in a standard firing profile. The resistance of as-fired thick-film structures was from the range 400– 500 X/sq for 100 X/sq resistor paste and 7–9 kX/sq for 10 kX/sq one. The Aurel NAVS-30 laser trimming system with Nd:YAG laser crystal was used for resistor trimming
353
Sheet resistor
Nanosecond Switch
microcontroller
Ohm -Meter
Trimming Data Base
HV-Relay
Fig. 1. Concept of high voltage discharge trimming and measuring system.
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359
for high ohmic resistors trimming. Therefore during trimming of 10 kX/sq resistors the 900 V voltage was kept constant but the duration of discharge time changed in the range from 5 to 70 ls. Increasing of pulse width means increase of electrical energy, which affects the resistance of the tested component. This means that the level of resistance changes after HV trimming was dependent on the strength of electric field in the case of low resistive components or the pulse duration in the case of high resistive ones. One should note, that in all of these investigations the resistors were unglazed, because we wanted to investigate the microstructures before and after trimming. LTCC resistors were buried in DP 951 Du Pont green tape. The 2 · 1, 1 · 1 or 1 · 2 mm2 resistors with Aubased terminations were made from three different resistor ink systems with sheet resistance from 10 X/sq to 10 kX/sq and fabricated according to tape manufacturer recommendation. Part of these resistors were subjected to high voltage pulse trimming (the trimming conditions were similar as for surface resistors). The trim target for these components was a resistance change of about 25%.
0
o
162 C
o
207 C
o
253 C
-4 ∆R/Ro [%]
354
-8 -12 -16 100 ohm/sq. -20
0
S2 - 1x1 mm2 S3 - 1x1 mm2 S5 - 1x1 mm2 S10 - 0.5x0.5mm2 100 200 300 2 S11 - 0.5x0.5mm S13 - 0.5x0.5mm2
400 500 t [h]
600
700
Fig. 2. Long-term stability of high voltage discharge trimmed 100 X/sq thick-film resistors (changes of resistance in trimming process: S2; +12.1%, S3; +6.0%, S5; +18.0%, S10; +3.1%, S11; +10.5%, S13; +19.0%).
high voltage pulse trimmed (HV) 162 °C
0
3. Long-term thermal ageing
207 °C
-1
253 °C
-2
∆R/ Ro [%]
Most often long-term thermal stability of thick-film resistors is characterized by the drift of resistance. Usually such a drift is characteristic for particular resistor systems, is dependent on ageing temperature and time and also on trimming method. But there are contradictory information about long-term thermal stability of high voltage trimmed resistors. Very good properties were reported in [7,8,11] but data presented in [10] have shown that 1 h keeping at 300 C created 3% relative resistance changes for pulse voltage trimmed samples and only about 0.1% for untrimmed ones. Investigations described in [14] have shown that long-term stability of HV exposed structures deteriorates with increase of number of pulses and is somewhat worse in comparison with laser trimmed or untrimmed ones. Similar conclusions were presented in [12]. Herewith we reported comparison of resistance drift induced by long-term thermal ageing at three different temperatures 162 C, 207 C and 253 C. The samples were kept at every temperature for about 200 hours. The relative resistance changes of high voltage discharge trimmed resistors are shown in Figs. 2 and 3. Long-term stability of untrimmed and laser-trimmed resistors is presented in Fig. 4. It is shown that untrimmed and laser-trimmed resistors are much more stable than voltage pulsed ones by extreme temperature stress. Untrimmed resistors exhibit resistance increase at all temperature levels. This increase is larger at the beginning of every temperature step but even after keeping the samples at 253 C it does
-3 10 kohm/sq. S2 - 1x1 mm2 S3 - 1x1 mm2 S5 - 1x1 mm2 S10 - 0.5x0.5 mm2 S12 - 0.5x0.5 mm2 S13 - 0.5x0.5 mm2
-4 -5 -6 0
100
200
300 400 t [h]
500
600
700
Fig. 3. Long-term stability of high voltage discharge trimmed 10 kX/sq thick-film resistors (changes of resistance in trimming process: S2; 13.0%, S3; 10.3%, S5; 5.4%, S10; 3.8%, S12; 10.5%, S13; 13.6%).
not exceed 1%. Very small laser trimmed resistors (0.25 · 0.25 mm2) have negative resistance drift after keeping at 207 C and 253 C but still within the range of ±1%. Resistors trimmed by energy of high voltage pulses show decidedly larger negative drift of resistance. This drift is much higher for unglazed 100 X/sq than 10 kX/sq structures and it depends on trimming level. Resistors with higher trim level are much less stable at every investigated temperature (but for applications up to 100 C the HV-pulse trimming method leads to practically acceptable results [8–11]). Moreover one should remember that this method is the only trimming
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359
untrimmed (UN) laser trimmed (L)
10 kohm/sq. 0.6
0.25x0.25 mm2 (UN) 0.25x0.25 mm2 (L) 1x1 mm 2 (UN) 1x1 mm 2 (L)
0.2 0.0
10 kohm /sq.; 1x1 mm2
1.02
R/R25
0.4
∆R/Ro [%]
1.03
o
untrimmed laser trimmed HV trimmed
1.01
1.00
162 C -0.2
207 oC
253 oC
0.99
-0.4
-150 -100
-50
(a)
-0.6
0
100
355
200
300
400
500
0 50 o T [ C]
100
150
600
t [h]
possibility for buried resistors. The explanation of this behaviour could be, that during the thermal ageing up to 253 C the conductive particles of high voltage trimmed resistors could crystallize and decrease their resistance value. More detailed structural investigations are necessary to explain completely such behaviour. This method is the only trimming possibility for buried resistors. For applications up to 125 C the HV-pulse trimming method leads to practically good acceptable results [8–11].
4. Normalized temperature dependence of resistance Resistance measurements of resistors in a wide temperature range are important for the analysis of conduction process. The temperature characteristics were measured automatically in the range from 170 C to 130 C. The Keithley 2000 multimeter interfaced to PC was used for data acquisition and presentation. The examples of normalized temperature dependence of resistance are shown in Fig. 5(a) whereas Fig. 5(b) presents differential temperature coefficient of resistance dR (TCRdiff ¼ RdT ) versus temperature. The results for 100 X/sq structures are qualitatively similar. In practice there are no differences between R(T) and TCR(T) characteristics for untrimmed and laser trimmed resistors. It should be noticed that HV-trimmed resistor has R(T) curve with minimum of resistance shifted significantly toward higher temperatures. The second difference appears above 100 C, where instead of expected resistance increase a noticeable drop of R is observed. The shift of a minimum on R(T) curve can be explained basing on the model proposed by Kozlowski [19]. He analysed the changes of the percolation path caused by appearing microcracks in thick-film resistors. The occurrence of the microcracks manifests itself as
TCR [ppm/K]
Fig. 4. Long-term stability of untrimmed and laser-trimmed 10 kX/sq thick-film resistors.
2
10 kohm/sq.; 1x1 mm
100
untrimmed laser trimmed HV trimmed
0
- 100 - 200 - 300 -150
-100
(b)
-50
0 o T [ C]
50
100
150
Fig. 5. Comparison of (a) normalized temperature dependence of resistance, (b) differential temperature coefficient of resistance for untrimmed, laser-trimmed and high voltage-trimmed 10 kX/sq thick-film resistors.
removing of certain bonds, randomly chosen from the initial critical percolation path. After such an operation a new percolation path is found and because of slightly different temperature dependence of a new ‘‘unit cells’’ in the percolation path the changes of the temperature dependence of resistance are induced. Kozlowski has shown both numerically and experimentally that the increase in density of microcracks caused a shift of the resistance minimum to the higher temperatures i.e. in the same manner as in Fig. 5(a). At this very moment there is no explanation for the curve bending at about 100 C. However this temperature may be important from the long-term stability point of view. Fig. 5(b) indicates the possibility of TCR shifting in high value resistors which can be useful and desirable in some applications.
5. Low-frequency noise Current noise index (CNI) is one of the most important parameters of resistors. But the information stored
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359
CNI ¼ 10ðlogðS ln 10Þ þ 12Þ
ð1Þ
An interesting observation is that the slope of logSV versus log f plots in Fig. 6(b) increases after the resis-
10 kohm/sq.; 1x1 mm2; HV(-4,71%); before ageing
10
13.73 V 11.16 V 8.59 V 5.15 V 2.59 V
-11
-12
2
S u [V /Hz]
10
10 10 10
-13
-14
-15
1
10
(a)
100 f [H z]
10 kohm /sq.; 1x1 mm2; HV (-4,71%); after ageing
10
2
in its value, sufficient for practical applications, is not sufficient to identify sources generating the fluctuations. For these purposes the measurements of power spectral density are necessary [20, and references therein]. The standard procedure of such low frequency noise spectroscopy is to examine the product fSV of frequency f and power spectral density SV of noise voltage. The common 1/f noise in fSV versus f plots appears as a flat background, whereas the features are associated with other kind of noises. Noise spectra of thick-film resistors usually possess a typical 1/f shape. Recently it was proven that also polymer thick-film resistors [21] as well as LTCC resistors both buried and surface [22] show Gaussian 1/f noise below 1 kHz. There are also papers [e.g. 23,24] where low-frequency noise measurements are used as a diagnostic tool for thick-film resistor evaluation. Even recently it was shown that low-frequency noise is more sensitive to the resistor degradation due to high-voltage pulse stressing than resistance itself [25]. However, this was shown for the first time and the number of experimental results is up to date not too large. In this paper we have measured and compared the low-frequency noise of untrimmed, laser trimmed and voltage trimmed 100 X/sq and 10 kX/sq thick-film resistors with 0.5 · 0.5 mm2 and 1 · 1 mm2 planar dimensions. After the measurements the resistors were thermally aged at 162 C for 170 h and then again subjected to noise measurements. Noise measurements were carried out by a standard DC technique. The sample of resistance R was biased through the ballast wire-wound resistor RB from a DC source. The noise signal was AC-coupled to the lownoise preamplifier Unipan 233-7 (bandwidth 0.5 Hz–100 kHz). Then it was sent to the HP 35660A digital signal analyser in which power spectral density SV of voltage dV was calculated in the frequency range from 2 Hz to 800 Hz for several voltages biasing the sample. After subtracting the noise background, SV=0 which was the sum of preamplifier noise and thermal noise, it has occurred that the spectra have 1/f shape (the line in log–log plot has a slope of approximately 1) (Fig. 6), Consequently fSV plots depend on frequency rather weakly (Fig. 7) and so, we use the average h fSViover the measured frequency range to estimate the frequency independent noise level. This quantity is still bias dependent. When logh fSVi is displayed versus logV (Fig. 8) the lines drawn through the data have the slope of approximately 2. This means that the relation SV V2 holds and the physical origin of the noise are equilibrium resistance fluctuations. The quantitative measure of such a noise is the ratio S = h fSVi/V2, which is related to CNI via the relation
Su [V /Hz]
356
10 10 10 10
11.89 V 8.93 V 5.96 V 2.92 V
-10
-11
-12
-13
-14
1 (b)
10
100 f [Hz]
Fig. 6. Power spectral density of ‘‘excess’’ low-frequency noise for high voltage trimmed 10 kX/sq thick-film resistor: (a) astrimmed (before ageing), (b) after thermal ageing (170 h, 162 C).
tors are thermally aged, and the plots in Fig. 7(b) are no longer flat. We leave the reasons for this behaviour for further studies. At this moment we summarize the effects of long-term thermal ageing on resistance drift and low frequency noise only on in a quantitative manner. This is done in Table 1. Data for low frequency are expressed in the terms of both power spectral density and current noise index. The changes in CNI are also shown in Fig. 9. It is interesting that the noise index of as-fired laser or HV-trimmed resistors is not worse than that of untrimmed resistors. After ageing we observe some insignificant variations of CNI for untrimmed or laser trimmed resistors. For HV-trimmed resistors, however, ageing process leads to noticeable increase of noise index. The increase of CNI (S) is larger in case of significantly trimmed structures (it was shown in Section 3, that HV-trimmed resistors with higher trim level are much less thermally stable). Let us note that while ageing the resistance of these (i.e. HV-trimmed) resistors has decreased by around 0.5–2%.
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359 2
2
10 kohm/sq.; 1x1 mm ; HV (-4,71%); before ageing
-10
2
f Su [V ]
10
-10
10
-11
B Linear Fit of Data3_B ###
2
13.73 V 11.16 V 8.59 V 5.15 V 2.59 V
f Su [V ]
10 kohm/sq.; 1x1 mm ; HV (-4,71%); before ageing
10
357
α =1.96
-11
10
10
-12
2
-12
10
1
10
4
6 U p [V]
(a)
100
8
10 12 14
f [Hz]
(a)
2
10 kohm/sq; 1x1 mm ; HV (-4,71%); after ageing
2
10 kohm/sq.; 1x1 mm ; HV (-4,71%); after ageing
2
f Su [V ]
-9
2
-9
10
10
f Su [V ]
11.89 V 8.93 V 5.96 V 2.92 V
-8
10
10
-10
10
-11
α =1.99
-10
10
10
-11
2
4
(b)
1 (b)
10
6 Up [V]
8
10 12 14
100 f [Hz]
Fig. 7. The product of frequency and power spectral density calculated for the data from Fig. 6(a) and (b).
6. Trim results and stability of buried LTCC resistors Long-term stability investigations were made after high-voltage-pulse trimming and after pre-conditioning. The latter was intended to simulate assembly processes.
Fig. 8. The dependence of averaged product of frequency and power spectral density on biasing voltage, fS V U aP (data for high voltage trimmed 10 kX/sq resistors) (a) before (b) after thermal ageing (170 h,162 C).
It contained 2· reflow soldering and next glob top curing for 5 h at 150 C. The final qualification tests are high temperature storage (1000 h, 125 C) and temperature cycling (1000 cycles from 55 C to 125 C).
Table 1 The effect of trimming method and ageing process on resistance drift and changes of relative power spectral density (S) and current noise index (CNI) Dimension (mm2)
Exposure Untrimmed Untrimmed Laser trimmed (+18.85%) High voltage pulse trimmed (1.9%) High voltage pulse trimmed (4.2%) High voltage pulse trimmed (4.7%)
Unaged Aged Unaged Aged Unaged Aged Unaged Aged Unaged Aged Unaged Aged
1·1 1·1 1·1 1·1 1·1 1·1
R (kX) 11.476 11.538 12.008 12.045 11.248 11.299 11.510 11.472 10.920 10.801 10.905 10.744
S(f=10 Hz)
CNI (dB) 14
2.88 · 10 4.32 · 1014 1.03 · 1013 8.07 · 1014 2.19 · 1014 1.76 · 1014 3.22 · 1014 6.03 · 1014 4.69 · 1014 2.05 · 1013 3.34 · 1014 5.50 · 1013
11.8 10.0 6.2 7.3 13.0 13.9 11.3 8.6 9.7 3.3 11.2 1.0
358
A. Dziedzic et al. / Microelectronics Reliability 46 (2006) 352–359
.
.
.
.
frequency noise) and stability (relative resistance drift, changes of current noise index after ageing) properties of untrimmed, laser trimmed as well as high voltage trimmed thick film resistors. Moreover they were correlated with additional resistor high temperature ageing after trimming process. According to our knowledge this is the first so wide comparison of various trimming method. During these investigations the following was found:
Fig. 9. Changes of current noise index of 10 kX/sq as a function of trimming method and ageing process.
After pre-conditioning the first results of trim stability were obtained. Contrary to components on alumina trimmed buried LTCC resistors drifted always in the trimming direction i.e. they showed the positive drift if resistance increased after trimming increase and the negative drift if resistance decreased after this process. One candidate (A-100) changed values of more than 5%. Two other pastes, having mean deviations of about 1%, had one or two group members exceeding this value up to 6%. The control groups showed very low deviations. Table 2 shows that some of the used resistor systems are suitable for high voltage trimming. Some of them (C-10k, B-10k, C-100) have a deviation < 1% after preconditions and thermal ageing and others (A-10, A-100, A-1k) have a drift up to 6%. This observation points out that the trimming stability after thermal ageing is depended on resistor paste. 7. Conclusions In this paper we presented wide spectrum of electrical (normalized temperature dependence of resistance, low
• There is distinct difference in long-term thermal stability of investigated components. The drift of untrimmed and laser trimmed resistors is similar. But structures trimmed by series of voltage pulses after extreme temperature stress have stability worse by about one order of magnitude. The drift is larger for resistors more trimmed by pulses (having larger resistance difference before and after trimming). • The temperature dependences of resistance in a wide temperature range are similar for untrimmed and laser trimmed components. However the minimum of resistance in HV-trimmed resistors is shifted to the higher temperatures. Moreover unexpected change of R(T) curve shape appears near 100 C. • The noise with a typical 1/f shape caused by an equilibrium resistance fluctuations is dominant in the low frequency range independently of the trimming method. The initial (i.e. before ageing) noise intensity is comparable for all resistors made of the same ink and of identical planar dimensions. After long-term thermal ageing the current noise index of HV-trimmed resistors noticeably increases in spite of simultaneous slight decrease of resistance. The increase of CNI is larger in case of strongly trimmed structures. On the other hand the noise level of untrimmed and laser trimmed resistors does not undergo changes after ageing process (similarly as resistance). • Trimming of unglazed resistors by energy of high voltage pulses induces other ageing mechanisms than those, which are predominant in untrimmed or laser trimmed resistors, which is essential for high temperature load. Such trimming method is useful in many standard applications where the temperature of resistors do not exceed 100 C.
Table 2 Long-term stability of LTCC resistors versus applied resistive ink Paste
Rsq after three refirings and tolerance
Trim range and direction of resistance changes
A-10 A-100 C-100 A-1k B-10k C-10k
50 X/sq ±50% 40 X/sq ±50% 80 X/sq ±20% 130 X/sq ±20% 2.3 kX/sq ±25% 3 kX/sq ±30%
>25% >25% >25% >25%
