SBOA533 January 2022 INA138 , INA138-Q1 , INA139 , INA139-Q1 , INA168 , INA168-Q1 , INA169 , INA169-Q1 , INA170 , INA180 , INA180-Q1 , INA181 , INA181-Q1 , INA183 , INA185 , INA186 , INA186-Q1 , INA190 , INA190-Q1 , INA191 , INA193 , INA193A-EP , INA193A-Q1 , INA194 , INA194A-Q1 , INA195 , INA195A-Q1 , INA196 , INA196A-Q1 , INA197 , INA197A-Q1 , INA198 , INA198A-Q1 , INA199 , INA199-Q1 , INA200 , INA200-Q1 , INA201 , INA201-Q1 , INA202 , INA202-Q1 , INA203 , INA203-Q1 , INA204 , INA205 , INA206 , INA207 , INA208 , INA209 , INA210 , INA210-Q1 , INA211 , INA211-Q1 , INA212 , INA212-Q1 , INA213 , INA213-Q1 , INA214 , INA214-Q1 , INA215 , INA215-Q1 , INA216 , INA2180 , INA2180-Q1 , INA2181 , INA2181-Q1 , INA219 , INA2191 , INA220 , INA220-Q1 , INA223 , INA225 , INA225-Q1 , INA226 , INA226-Q1 , INA228 , INA228-Q1 , INA229 , INA229-Q1 , INA2290 , INA230 , INA231 , INA233 , INA234 , INA236 , INA237 , INA237-Q1 , INA238 , INA238-Q1 , INA239 , INA239-Q1 , INA240 , INA240-Q1 , INA240-SEP , INA270 , INA270A-Q1 , INA271 , INA271-HT , INA271A-Q1 , INA280 , INA280-Q1 , INA281 , INA281-Q1 , INA282 , INA282-Q1 , INA283 , INA283-Q1 , INA284 , INA284-Q1 , INA285 , INA285-Q1 , INA286 , INA286-Q1 , INA290 , INA290-Q1 , INA293 , INA293-Q1 , INA300 , INA300-Q1 , INA301 , INA301-Q1 , INA302 , INA302-Q1 , INA303 , INA303-Q1 , INA3221 , INA3221-Q1 , INA381 , INA381-Q1 , INA4180 , INA4180-Q1 , INA4181 , INA4181-Q1 , INA4290 , INA901-SP , LM5056A , LMP8278Q-Q1 , LMP8480 , LMP8480-Q1 , LMP8481 , LMP8481-Q1 , LMP8601 , LMP8601-Q1 , LMP8602 , LMP8602-Q1 , LMP8603 , LMP8603-Q1 , LMP8640 , LMP8640-Q1 , LMP8640HV , LMP8645 , LMP8645HV , LMP8646 , LMP92064
The results given in Section 2 initially imply that copper trace shunt resistors are not feasible for practical use, given the inability to control the true thickness of the trace and tendency of copper to change resistance as current flows through it. However, Figure 3-1 displays a second design revision with an alternative experimental setup that could demonstrate how to avoid the issues discussed previously. This design is referred to as Revision B.
The trace at the top (see Figure 3-1) attempts to add a large amount of surrounding copper (an extended ground plane) to the 100-mil trace to decrease the impact of the PCB manufacturing error. The second trace is a normal 100-mil trace that is measured by both the INA190, as in the first revision, and the INA181. The INA181 current-sense amplifier is designed for cost-optimized applications. This device is part of a family of bidirectional, current-sense amplifiers (also called current-shunt monitors) that sense voltage drops across current-sense resistors at common-mode voltages from –0.2 V to +26 V, independent of the supply voltage. The INAx181 family integrates a matched resistor gain network in four fixed-gain device options: 20 V/V, 50 V/V, 100 V/V, or 200 V/V. This matched gain resistor network minimizes gain error and reduces the temperature drift. The reason for comparing the INA181 to the INA190 is to analyze the possibility of using a copper trace shunt in conjunction with the lower cost INA181 to develop a less expensive but less accurate current-sense solution. For this trace, no attempt is made to regulate the thickness of the trace. Instead, a two-point calibration is used to try to accurately predict the output of the device regardless of what its actual thickness is. This kind of calibration is also tested with other trace widths. Finally, the last three traces in Figure 3-1 are repeated 8-mil traces, intended primarily to look at the variability in trace width within a single board.
Table 3-1 shows the percent error results for the 100-mil trace with an extended ground plane. The 100-mil trace errors from Table 2-1 are reprinted for comparison, as well as the 100-mil trace with no ground plane shown in Figure 3-1. As before, “board 1” and “board 2” refer to different boards in the same revision.
| Trace | Average Percent Error | Average | ||
|---|---|---|---|---|
| 1” | 2” | 3” | ||
100 mil extended ground plane, board 1 | –16.71% | –20.29% | –23.22% | –20.07% |
100 mil extended ground plane, board 2 | –18.25% | –21.43% | –24.64% | –21.44% |
100 mil, board 1 | –38.87% | – | – | –38.87% |
100 mil bottom tap off | –42.96% | –44.08% | –48.38% | –45.14% |
100 mil center tap off | –38.53% | –39.25% | –39.15% | –38.98% |
The error is significantly reduced even when compared to a trace that is on the same board, but is still large. However, based on the results of testing multiple boards, it appears that the error is at least consistent. This could mean that this technique is feasible, but it is also possible that a different board from a different manufacturer would have a different error. In addition, the larger ground plane takes up a significant amount of space and essentially removes the advantages of using a smaller 100-mil trace as opposed to the 1750-mil trace, as the latter option is more accurate and occupies approximately the same area. The results of this experiment indicate that the more continuous the copper plane, the closer trace thickness is to the expected value and that obtaining a resistance of the correct value would require a very large trace.
The board was also used to determine the effectiveness of a simple calibration process which would save space. The curves shown in Figure 2-1 indicate that the discrepancy between expected and actual outputs could simply be treated as a gain error. Calibrating the trace with a low and a high current could theoretically allow for any outputs along the Actual curve to be predicted. This procedure is complicated by several factors. First, as shown in Figure 2-4, resistance changes as current moves through the trace. This means that the calibration curve can be skewed depending on when measurements are taken. In some cases, outputs settling times were recorded in excess of 5 minutes. Also, if there is significant variability of trace thickness between boards, calibration would be required for every board and a batch calibration process would suffer from inaccuracies. Finally, calibrating in this way does not allow for changes caused by temperature variation away from the calibration temperature.
To determine the feasibility of a two-point calibration method, the first step was to use a realistic procedure. Table 3-2 displays 4 possible setups, distinguished by how many data points were collected, whether or not the calibration output measurements were allowed to stabilize, and whether or not the test output was allowed to stabilize. Each setup was used to predict the output of the INA190 output with 2.5 A running through the trace. The percent error between this prediction using the calibration and the actual output is also given. These were obtained with the plain 100-mil trace in Figure 3-1.
| Setup | Number of Data Points | Calibration Outputs Stable? | Test Output Stable? | Percent Error |
|---|---|---|---|---|
Max. Data Points | 4 | Yes | No | –1.85% |
Reduced Data Points | 2 | Yes | No | –1.24% |
Reduced Temp. Effects | 2 | No | No | –0.58% |
Max. Temp. Effects | 2 | No | Yes | –2.25% |
The first setup took four calibration data points, three of which were in the low current range. This was the most unrealistic, as in a practical application it would most likely not be feasible to wait for four different calibration points to stabilize before recording them. Also, a real application would most likely always be dealing with a current that had been running through the trace long enough to bring it up to its equilibrium temperature, so recording the test output before it stabilizes is not the most accurate simulation of actual events. For this reason, the second and third setups are also not realistic. The fourth option represents what would probably be implemented: the minimum number of data points, with the shortest amount of time spent on waiting for the calibration process, and a test current that has been flowing for a long time. Unfortunately, this procedure has the largest error but is the only one that could realistically be implemented in large quantities.
To test the calibration process, the data points recorded were used to calculate the slope and intercept of the calibration curve. With these numbers, it is possible to backwards-calculate an output from the INA190 and predict the current. Because the actual current is known, calculating percent error reveals the effectiveness of the calibration. The first trace that was calibrated was the 100-mil trace used to obtain Table 3-2 data. The calibration currents used were 0.1 A and either 5 A or 10 A. Four test currents were used: 0.01 A, 2.5 A, 5.5 A, and 7 A. For each, the INA190 output was allowed to stabilize before recording. To thoroughly examine the capabilities of this technique, the 100-mil trace from revision B was used to predict INA190 outputs from the 200-mil trace of revision A. Table 3-3 and Table 3-4 show the results. The calibration data was taken from board 1, revision B.
| Board | Max. Cal. Point | Percent Errors | |||
|---|---|---|---|---|---|
| 0.01 A | 2.5 A | 5.5 A | 7 A | ||
Board 1, Rev. B | 5 A | –36.31% | –1.16% | 2.11% | 4.03% |
Board 2, Rev. B | 5 A | –12.72% | –1.58% | 1.46% | 3.53% |
Board 1, Rev. A | 5 A | –98.76% | –1.87% | 1.98% | 4.55% |
Board 1, Rev. B | 10 A | –20.52% | –4.57% | –1.49% | 0.35% |
Board 2, Rev. B | 10 A | 15.20% | –4.97% | –2.12% | –0.14% |
Board 1, Rev. A | 10 A | –81.22% | –5.25% | –1.62% | 0.85% |
| Board | Max. Cal. Point | Percent Errors | |||
|---|---|---|---|---|---|
| 0.02 A | 5 A | 10 A | 20 A | ||
Board 1, Rev. A | 5 A | –79.87% | 18.01% | 24.96% | 49.92% |
Board 1, Rev. A | 10 A | –55.87% | 13.92% | 20.54% | 44.57% |
Testing two different calibration points demonstrates how the calibration can be adjusted based on anticipated current. A 100-mil trace can handle 5 A while staying within a relatively arbitrary limit of 20°C of temperature rise. Calibrating to a higher current allows for more accurate predictions of larger currents, but the extrapolation of lower currents suffers. Also, accuracy of the calibrated prediction severely suffers when used for a trace of a different width and revision. This makes sense as more factors are introduced that can cause deviation from the calibration conditions.
This calibration procedure was also used for the trace with the extended ground plane, as well as the 8-mil trace. Table 3-5 and Table 3-6 show these results. The two calibration points used for the 8-mil trace were 0.02 A and
1 A.
| Board | Max. Cal. Point | Length | Percent Errors | |||
|---|---|---|---|---|---|---|
| 0.01 A | 2.5 A | 5.5 A* | 9.5 A* | |||
Board 1 | 5 A | 1 in | 800.08% | 3.00% | 4.57% | 10.73% |
2 in | 65.48% | –0.64% | 2.29% | – | ||
3 in | –28.81% | 11.92% | – | – | ||
Board 2 | 5 A | 1 in | 796.22% | 2.47% | 3.63% | 10.24% |
2 in | 81.88% | –1.63% | 0.75% | – | ||
3 in | –29.23% | 10.39% | – | – | ||
| Board | Percent Errors | |||
|---|---|---|---|---|
| 0.005 A | 0.5 A | 0.95 A | 1.5 A, 1.2 A* | |
Rev. B, Board 1, 1” | 12.99% | –0.72% | 0.89% | 3.51% |
Rev. B, Board 2, 1” | 4.06% | –2.03% | –0.45% | 2.00% |
Rev. B, Board 2, 3” | 9.91% | 1.59% | 2.71% | 3.58% |
Rev. B, Board 1, square | –1.85% | –14.46% | –12.97% | –11.69% |
Rev. A, Board 1, square | –9.70% | –8.71% | –7.10% | –5.83% |
Finally, the INA181 was used for calibration. The procedure was identical to the method used for the INA190, only with a different device. Only the 100-mil trace was examined using this calibration technique. Table 3-7 shows the results.
| Board | Max. Cal. Point | Percent Errors | |||
|---|---|---|---|---|---|
| 0.01 A | 2.5 A | 5.5 A | 7 A | ||
Board 1, Rev. B | 5 A | –99.90% | –0.12% | 2.24% | 4.47% |
Board 1, Rev. B | 10 A | –99.90% | –2.93% | –0.71% | 1.44% |
Board 2, Rev. B | 5 A | –52.27% | –0.89% | 1.61% | 3.66% |
Board 2, Rev. B | 10 A | –27.50% | –3.68% | –1.32% | 0.66% |
The data in the previous tables demonstrate several limitations of the two-point calibration process. Calibration points taken from one board usually were able to predict outputs from boards of the same revision. However, for boards of different revisions, the accuracy was significantly decreased due to PCB manufacturing variability over time. The INA181 also seemed to perform equal to or even better than the INA190. This suggests that it can be used in place of the INA190 with similar results, at least when calibration is being used to account for discrepancies.
Finally, revision B provided the opportunity to reexamine the effects of trace shape on resistance. Table 3-8 shows the results from these experiments, similar in format to Table 2-1.
| Trace | Average Percent Error | Average | ||
|---|---|---|---|---|
| 1” | 2” | 3” | ||
8 mil | –51.95% | –53.64% | –53.20% | –52.93% |
8 mil square left | – | – | –58.01% | –58.01% |
8 mil square right | – | – | –51.80% | –51.80% |
While there are differences, it is difficult to determine whether these are due to any impact from the shape of the trace or simply because the thicknesses are different. The fact that the same shaped trace has a different average (right versus left) implies that this discrepancy is due to the same previously-discussed tolerances, or at least that any difference in resistance added by trace shape is not enough to overcome thickness variation.
All results showed a much larger percent error for very low current values, but this is to be expected due to the offset error of the INA190 and INA181. These would most likely be of concern even for a conventional SMT resistor.