Testing and Characterization of the MPA Pixel Readout ASIC for the Upgrade of the CMS Outer Tracker at the High Luminosity LHC Dena Giovinazzo University of California, Santa Cruz Supervisors: Davide Ceresa and Kostas Kloukinas August 2018
Giovinazzo 1 Abstract The Macro Pixel ASIC (MPA) is a readout chip designed for the CMS Outer Tracker upgrades at the High Luminosity LHC (HL-LHC). The MPA s basic functionalities have already been tested and new research has been done on potential optimizations and the MPA s response to different stimuli. This paper will examine the effect of the krummenacher feedback and preamplifier bias on the analog performance of the MPA, analyze the response of the MPA when it detects particles from a radioactive source, verify that the process of bump bonding the MPA wafer does not damage the functionality of the chip, and test the radiation hardness of the chip. Table of Contents Abstract 1 Introduction 2 Tests and Results 3 Krummenacher and PreAmplifier 3 Response from Radioactive Source 8 Wafer Probing 12 Irradiation 14 Conclusion 17 Acknowledgements 18 References 18
Giovinazzo 2 Introduction 1 The HL-LHC upgrade will increase the number of collisions by a factor of 10. This will increase the amount of data that must be readout, triggering a new readout ASIC design. The MPA was designed to not only quickly readout events, but also select only interesting events to pass to the back end electronics to limit the already massive amount of data being transmitted. Initial testing on the MPA proved it met initial parameters, but more specific tests were needed to observe its functionality in other scenarios. The MPA has seven DACs that can be set to change its performance. Two of these DACs, the krummenacher feedback bias and preamplifier bias, were thought to have some influence over the analog performance of the MPA. The krummenacher feedback provides compensation for leakage current and changing the amount of compensation was expected to have an effect on the gain of the MPA with respect to the calibration value, due to the krummenacher feedback changing the peak of the signal. It was also expected that increasing the feedback will decrease the noise response. The combination of these responses is the goal of testing the krummenacher feedback. The current allocated to the preamplifier can also be controlled. Power consumption can be significantly decreased by providing less current to the preamplifier, but how this affects the performance of the MPA must be determined. Previous MPA tests used an injected charge to observe the response of the chip. The injected comes from a capacitor, which slightly alters the rest of the system. This capacitor will not be used in the experiment, so the performance of the MPA must be evaluated when the capacitor is disabled and the signals come from actual particles. A radioactive source within the dynamic range of the MPA front-end can provide an absolute characterization. Like all ASICs, MPA chips are produced on a wafer. The chips are tested after every stage of production. This process includes the manufacturing of the active wafer, bump bonding, and dicing. The wafers contain 88 chips that must be individually tested after each step. This is done by a wafer probing machine, which can be programmed to test each chip on the wafer and is used for this experiment to test the functionality of the wafer after being bump bonded. The MPA will be in a high radiation environment during the course of the HL-LHC experiments. It is important that it can function properly during this time without failing due to radiation. Putting the MPA in an X-Ray machine will test its radiation hardness by exposing it to a high dose rate. 1 http://hilumilhc.web.cern.ch/
Giovinazzo 3 Tests and Results Krummenacher and PreAmplifier The testing procedure for the krummenacher and preamplifier analysis is as follows: The chip was first calibrated and trimmed. The krummenacher DAC value was set to 0 and stepped by 1 to 31. A row-by-row injection s-curve was done for every krummenacher value with the calibration set to 20. The gain was also found for two different krummenacher DAC values. First, the DAC value was set to 6 and two s-curves were run; one with a calibration of 15 and the other with a calibration of 30. This was repeated with the krummenacher DAC set to 24. The gain was calculated for each individual pixel using the following formula: G = [T hresholdcal = 30 T hreshold cal = 15 ] * T hlsb CalLSB(30 15) [ mv /fc] T hreshold cal = 30 is the threshold value extracted from the s-curve when the calibration is set to 30. T hreshold cal = 15 is the same, but when the calibration is set to 15. ThLSB is the threshold LSB, which is calculated by setting the threshold DAC to 0 and 160, measuring the output, then computing the slope between the two points. The same is done for CalLSB, which finds the calibration DAC s LSB. Additionally, the typical LSB value of the MPA is 100 electrons. The results of this test show that as the krummenacher feedback is increased, the average threshold value of the pixels decreases exponentially, from ~94 LSB to ~87 LSB. The threshold spread also decreases exponentially from 6.9 LSB to 5.4 LSB. The noise decreases somewhat linearly from 2.7 LSB to 2.4 LSB. The current increases a negligible amount from 5.88 ma to 5.92 ma, which can essentially be ignored. The gain decreases by about 9 mv/fc, from just below 78 mv/fc to just above 68 mv/fc.
Giovinazzo 4 Figure 1. Graph of the average threshold as the krummenacher DAC changes from 0 to 31. Figure 2. Graph of the average threshold spread as the krummenacher DAC changes from 0 to 31.
Giovinazzo 5 Figure 3. Noise variation as the krummenacher DAC was increased from 0 to 31. Figure 4. Graph of the current draw as the krummenacher DAC changes from 0 to 31. Note that the three voltage values are intentionally disregarded.
Giovinazzo 6 Figure 5. Gain when the krummenacher DAC = 6 and 24. The same method of testing the krummenacher values was also used to test the preamplifier DAC. The results showed that increasing the preamplifier DAC had a small effect on the noise, decreasing it from 2.56 LSB to 2.49 LSB. Increasing the preamplifier DAC increased the power consumption considerably, as expected. The current draw increased from 50 ma to 65 ma. The average threshold spread seems to be unaffected by any change in the preamplifier DAC, indicating that the threshold is not affected by the preamplifier. And finally, the gain had no tangible effect from the preamplifier, decreasing by less than one mv/fc. Figure 6. Graph of the noise variation as the preamplifier DAC changes from 0 to 31.
Giovinazzo 7 Figure 7. Graph of the current draw as the preamplifier DAC changes from 0 to 31. Voltage not pictured. Note that the voltage is intentionally disregarded. Figure 8. Graph of the average threshold spread as the preamplifier DAC changes from 0 to 31.
Giovinazzo 8 Figure 9. Gain when the preamplifier DAC is set to 6 and 24. Based on these results, it is clear that the preamplifier DAC can be set to a low value with minimal effects on the performance of the MPA. The krummenacher DAC can be used to tune the threshold value. Noise decreases as the krummenacher value increases, but so does gain, and a balance must be found between the two. Response from Radioactive Source Americium-241 was used to see how the MPA would respond to detecting hits from actual particles. A test set up was configured to enable the readout of the pixels, but disable the injection capacitor which is used to inject a charge similar to a particle. S-curves were run where, at each calibration value, every pixel s shutter was open at the same time for 3-5 minutes. The first test used only Am-241, where we expected to see a threshold value of 220 LSB based on the typical LSB value of 100 electrons. The average threshold of the data turned out to be 200, with a spread of 3.98 LSB. Below is a heat map showing the number of particles counted for every pixel of the MPA. Pixels 1 and 2 have their outputs tied together which is why they show double the count. This is also true for pixels 119 and 120.
Giovinazzo 9 Figure 10. Heat map showing the particle count for every pixel. Figure 11. Threshold histogram of the AM-241 radiation test.
Giovinazzo 10 To see how the entire MPA responded, we summed up the counter value of every pixel for each threshold to get a sum s-curve plot, as shown below. Figure 12. S-curve of the sum of all pixel counts using Am-241. The blue line is the sum data, and the orange line is the curve fit. The curve fit produces a threshold of 201. Interestingly, the sum data is not flat before curving down. This is likely due to neighboring pixels injecting charge into each other, but this hypothesis requires further exploration. This phenomena means the noise of the system can not be extracted. The MPA was also tested with Am-241 irradiating tin above the sensor. This produced a second radioactive particle, which in turn produced a second s-curve.
Giovinazzo 11 Figure 13. S-curve of the sum of all pixel counts using tin and Am-241. The first s-curve is very close to the noise peak, which is why the value is so high at the beginning. This causes some residual noisiness between threshold DAC values 80 and 100, further complicating the extraction of the threshold. Analysis of this data is relient on the understanding of neighboring pixel charge carrying. Further explorations were done on this phenomena using charge injection instead of a radioactive source. The results showed that small amounts of charge carrying are seen when the entire row is injected at once, but just injecting neighboring pixels has no effect. During the wafer probe tests, described below, a row-by-row charge injection was also seen to cause a major increase in the noise of the pixels. From these results it is possible to calculate the absolute calibration of the MPA. By taking the energy peak of Americium and dividing it by the energy needed to produce electron-hole pairs in Si, the number of electrons produced by Americium can be found. Then, the number of electrons is divided by the difference in threshold LSB between the observed threshold and the noise peak, and the electron LSB, or absolute calibration of the chip, can be found. For the test using Am-241 and tin, the Am-241 threshold was found to be 200, and the noisepeak was guessed to be 70. The actual value is not known because the test started the Th DAC at 70. The energy peak of Am-241 is 60 kev and the energy to create electron-hole pairs in Si is 3.6 ev.
Giovinazzo 12 E N e = Am 60 [kev Ee = ] 6.7 [ke ] 3.6 [ev ] = 1 e LSB = N e T h Noise = 16.7 [ke ] 200 70 = 128 [e LSB] The measured LSB of this chip sas 128 electrons, which is 28 electrons higher than expected. This could be caused by mismatch and other non-idealities. The absolute calibration of the chip can then be used to calculate the number of electrons that contribute to noise. From the krummenacher test above, the noise when the DAC = 15 is 2.56. Taking this value and multiplying by the electron LSB gives 327 electrons produced by noise. Charge carrying creates a resolution limit for the MPA. If the amount of charge carrying can be calculated, then the resolution can be recovered. Currently this value is unknown. Additionally, the absolute calibration of the chip will require another round of tests to find a more accurate value rather than the estimation calculated above due to the unknown value of the noisepeak. Wafer Probing StatsChipPac provided a bump bonded wafer that needed to be tested. The wafer probe machine tests each chip individually for their analog and digital response. First, the chip is powered on and the MPA is enabled. The script checks the current draw to ensure it is powered on. If the chip does not turn on, the test immediately ends and moves on to the next chip. After this, it checks the alignment DAC values which should be close to 15. If they are close to 0 or 31, it indicates that there may be an error. It next performs a shift test which checks that all output lines have proper signal integrity. If the signal integrity is too poor, the signal can not be recovered from the noise floor and all measurements will be lost. Following the shift test is the s-curve measurement which extracts the gain, noise, threshold spread, and calibration and threshold DAC LSBs. After this, the analog and digital components of each pixel are tested individually. It tests that the analog part of each pixel returns the correct value, and then checks the digital memory of each chip when the digital voltage is at 1V and 1.2V. The memory was designed to work at 1.2V, but simulations showed that it should also work at 1V. Consequently, if the memory works at 1V, it is guaranteed to work at 1.2V. After testing the wafer, only one chip could not be powered on at all as it had a short. The alignment and shift tests were passed by all chips. All chips had an analog response, though three showed poor s-curve results. Seventy-one chips had no memory issues at 1.2V, but most had issues at 1V. This is a common occurrence for other wafers so it was an expected result.
Giovinazzo 13 Figure 14. Histogram showing the spread of threshold values for all 88 chips. Figure 15. Histogram with the average noise for all 88 chips.
Giovinazzo 14 The results of the wafer probe test were very similar to tests that had been run on non-bump bonded wafers. The bump bonding process, therefore, does not affect the performance of the chips. Irradiation The MPA will be in a high radiation environment while it is operating. As such, it is necessary to ensure it can operate in this condition for long periods of time. To accomplish this, the MPA was put into an X-Ray chamber where it was continuously irradiated for several days. Tests were run during this time to check the status of the chip. The tests were similar to the probe station tests, but tested every DAC point rather than just two. It was suspected that irradiation would affect the DACs so it was necessary to see how every point evolved over time. One MPA chip received over 200 MRad of radiation. Below is how the DACs changed with radiation. The 300 MRad point is not actually 300 MRad; it was taken when the test ended and but the chip was still in the X-Ray machine. Similarly, the 400 MRad point was also taken after the test ended, but was in the lab test bench set up. Figure 16. DAC values as a percentage of their original, non-irradiated value over dose amount. Around 150 MRad, the DACs all jumped by 5-10% with the exception of the calibration DAC. The reason for this is currently unknown, and further testing is required. Additionally, the DAC values all immediately decrease once irradiation starts, again with the exception of the
Giovinazzo 15 calibration DAC. Another test will need to be run, taking more data points between 0 and 10 MRad to show this jump. Another interesting result is threshold DAC, shown below. Figure 17. Threshold DAC at the beginning of radiation vs. after radiation was completed. The blue lines are the threshold DAC when the MPA was in the X-Ray machine, before irradiation had started. The magenta lines are the threshold DAC when the MPA was moved to the lab test bench after being irradiated. The blue lines show a discontinuity just before 150 MRad, which has never been seen outside of the X-Ray machine. The reason for this discontinuity must be further researched, but it is possible that this is caused by some electrical or magnetic property of the X-Ray machine. The threshold and noise response of s-curve measurements show a similar result to the overall DAC values jumping immediately when radiation starts.
Giovinazzo 16 Figure 18. Average threshold for each run as the dose increased. Figure 19. Threshold spread for each run as the dose increased.
Giovinazzo 17 Figure 20. Average noise as the dose increased. The 300 MRad and 400 MRad values are the same as described above. Once the MPA has passed 10-15 MRad, the threshold and noise values level out. The initial irradiation phenomena must be further explored in the future. This irradiation test proved that the MPA can operate successfully even after receiving high amounts of radiation. Some DAC values stayed the same even after irradiation, which was expected. Each aspect of the chip s operation continued to work throughout and after the irradiation. Conclusion The MPA continues to meet expectations as various tests stress the chip. Optimizations with the krummenacher and preamplifier DAC values should be further explored and implemented. Am-241 testing showed that the absolute calibration of MPA chips can vary significantly from the expected value. The first bump bonded wafer showed promising results, and future wafers can likely be safely bump bonded as well. The effects of radiation are present in the chip but not to a damaging extent.
Giovinazzo 18 Acknowledgements I would like to thank Phil Rubin and the IRES at CERN program for providing me with this opportunity, and the National Science Foundation for making this opportunity possible. Thanks to Davide Ceresa and Kostas Kloukinas for their guidance and support throughout this project. And a special thanks to my colleagues on floor 6 for making me a master table football player. References [1]. The HL-LHC Project. http://hilumilhc.web.cern.ch/ [2]. MPA-SSA Project. https://espace.cern.ch/cms-mpa/sitepages/home.aspx