Together with Micron personnel, I performed various characterization tests on the four detectors. Several of these tests are done on test structures that we designed around the perimeter of the detector, on otherwise unused silicon. In general, the character of a particular process step is uniform across a wafer, and so indicative results can be obtained from these test structures and the delicate detector pads do not have to be directly probed. Some tests, such as looking for shorted strips and measuring leakage current, have to be done on the detector.
The p-implants on the MVD detectors are biased through polysilicon resistors. The detector bias resistors are made by depositing a layer of polysilicon on the wafer and then doping the layer to give it the correct resistance. One wants to get large values of resistance for good noise isolation from the power supply, and in order to acheive this without wasting active detector area, the resistors are etched in a serpentine pattern. The bias is applied through an individual resistor to each of the 256 strips on the detector. We placed 16 test resistors, with identical geometry, on the wafer below the detector, equally spaced from strip 1 to strip 256. These allow us to determine the uniformity of the resistor doping across the detector. The design value for the resistor was specified to be 5M-ohm. The value that we measured ranged from 3.0M-ohm to 3.75M-ohm across all four detectors. The spread in resistor values across an individual detector, as measured on the test patterns, was on the order of 5%. Both numbers are acceptable and do not in any way compromise the usefulness of the detectors as prototypes. A 5M-ohm resistor will give slightly better noise performance. The Micron process team felt very confident that with the data point they have from these first-stage prototypes, that they will be able to get very close to 5M-ohm in the second stage production. A test pad of 10 squares of doped polysilicon is also provided on each wafer. The corresponding values in units of ohms-per-square ranged from 6 to 8 k-ohms-per-square.
A contact pad is formed at one end of the bias resistor to provide a low resistance connection to the aluminum trace that goes between the resistor and a common bias bus. The pad is formed by doping the area more highly than for the resistor. We measured this value on a test structure to be about 120 ohms-per-square. This value is acceptable.
There are also p-implant test structures on each wafer to measure the resistivity and hence dopant concentration in the p-layer. The measurements on these structures were cosistent around 95 ohms-per-square. This is acceptable.
The MVD strips are ac-coupled. To accomplish this the p-implant that forms the strips is covered by a thin oxide layer that runs the entire length of the strip. An aluminum ac electrode is then deposited along the entire length of the strip over the oxide. The p-strip - oxide - aluminum sandwich forms the capactitor. The value of the coupling capacitance should be as large as possible since the amount of charge collected will be proportional to the ratio of the coupling capacitance to the capacitance of the strip to the backplane of the detector. This implies that the oxide layer should be as thin as possible, however there is a tradeoff between maximizing the capacitance and minimizing the number of faults due to pinholes which occur in the thin oxide. Micron currently grows the coupling oxide to 2000 angstroms. We measured several test capacitors on each of the four wafers. The value for the short strips was 220 pf, and uniform amongst different test structures. The value for the long strips was 175 pf, and also uniform across the test structures. Both values are acceptable and are on order of 100 times the capacitance from the strip to the backplane.
Under normal operation there should be no dc potential across the coupling capacitor as both the bias on the srtip and the dc level at the electronics input is the same. However, in the condition where there is a high instantaneous flux in the detector (as in a beam spill) a significant percentage of the entire backplane potential can be placed on the strip. The backplane potential for the MVD might reach 100V. With this in mind, we measured the current vs voltage across test capacitors on each of the four wafers, stepping up the voltage in 5V steps to a maximum of 110V. At the maximum voltage, the current through the test capacitors did not exceed 3nA, indicating that the capacitors were not breaking down. This was acceptable.
We measured the depletion voltage of each detector by plotting the capacitance of the p-implant to the backplane of the detector, vs the bias voltage across the detector. The capacitance decreases as the detector becomes depleted, and the inflection point of the curve where it begins to flatten is the voltage at which the detector is fully depleted. We made these measurements on 1cm x 1cm test diodes on each of the four wafers. The four had typical C/V characteristics with the inflection point ocurring at about 10V. This is lower than the value we specified, but is acceptable. A low depletion voltage limits the lifetime of a detector in a high radiation environment which should not be a consideration for MVD given the projected total lifetime integrated flux at the detector location in PHENIX. The depletion characteristic is determined by the wafer stock material, which is purchased from outside Micron.
The flatband voltage of a test MOS structure was also measured on the wafers. This test measures the quality of the field oxide on the wafers. All four flatband characteristics indicated good quality oxide layers. This measurement is not part of our specifications, but is used in-house as a quality assurance parameter.
The total leakage current of the detectors under full bias was measured on each of the detectors, both on the bias line and on the guard ring. All four detectors had total current of roughly 300 nA on the bias, corresponding to an average value of just over 1 nA per strip. This is well within specification and acceptable. Three of the four had acceptable guard ring current, ranging from 60 to 100 nA. The fourth, however, had a guard ring current of nearly 9 uA, and because of this, this detector is not acceptable.
Each strip on all four detectors was measured to determine the number of shorts, or pinholes found in the coupling oxide. Across te four detectors the number of measured faults was 1,1,0, and 3. This is acceptable. Micron personnel performed this test before I arrived, and there was not enough time to redo it in my presence. This test will be performed again at LANL before the detectors are officially accepted.
I did a visual inspection of the detectors under a microscope. The detector that failed the C/V and the total current measurements also had a visual defect in the middle of the active area of the detector. It was not possible to characterize the exact nature of the defect, but it appeared to affect the passivation, the oxide and the metal layers on the detector.
I measured the implant width on both the inner and outer detectors. The width was 50 um, which is our specification. I also measured the aluminzation width, which was 40 um, also conforming to our specification. The aluminum should be slightly less wide than the implant to avoid microdischarge at the edges of the aluminization.
Micron should have, but did not document the coupling oxide thickness during the process run. I had them take a test structure from an inner and an outer wafer, and remove the passivation and metal from it in a sulphuric acid bath. This left the coupling oxide exposed. Through spectral analysis, Micron personnel were able to measure the thickness of the oxides. The inner detector had 2050 angstoms of oxide, while the outer detector had 2100 angstoms of oxide. This is considered uniform and meets the specification.
I spent a full day in front of the CAD workstation with Mcron designer Anthony Lucas. The first area we worked was the readout traces. We finalized the road width between adjacent pad rows for the traces to be 300 um. The trace width will be 12 um with 10 um gaps. The gap between the inner and outer traces to the respective pad rows will be 17 um. The corners of the traces, where they form right angles will all be rounded. The interface where the trace meets the pad electrode or bonding pad will be tear-dropped. These two changes decrease the fault probability. The pad diffusion corners are also rounded to improve the breakdown characteristics. The road between pad columns is 200 um.
There will be two guard rings on the detector. The inner guard ring surrounds the pads and is as close as possible to them. Because of the pad shapes, the distance from the guard to the bottom of the pad is 50 um, and 100um from the guard to the top of the pad. The alternative would involve a complicated guard structure, which I did not propose. The guard ring implant will be 50 um wide. The guard ring will be aluminized, wherever practical, to minimize its resistance..The bias bus for the pads will cross over the inner guard ring on top of field oxide layer. The bias line is 100 um wide and is connected to each pad through a polysilicon resistor. The resistors are located in the 300 um roads between adjacent pad rows. The serpentine line that forms the resistor is only 5 um wide. This necessitates the use of a non-standard electron beam spot size of 0.25 um (0.5 um is standard) for the mask production. This will cost Micron some extra money, but does not affect our cost. The output traces will cross over the inner guard ring on top of a field oxide layer. They will then terminate in 80 um x 200 um bond pads.
Between the bond pads and the cut edge of the detector I have placed a second, outer guard ring. The ring runs the entire periphery of the detector and is intended to block noise current that originates along the cut edge from getting onto the bond pads or the bias line. The bond pads to connect to the guard rings and the bias line are each 200 um x 2000 um to allow for multiple probes and bonds.
There are contact pads, called spy pads, to each of the p-implants for probing. I also added contact pads to the ac electrode of the pads. This is a very delicate area to probe given the 2000 amgstrom oxide beneath it, however the contact may be useful in tracing or repairing certain types of faults.
There are several technical difficulties associated with the production of double metal detectors. One of the problems that Micron encountered was in the formation of the via holes that connect the first metal layer to the second. They would etch the hole based on a time, and the results were inconsistent, resulting in a large number of shorts between metal two and the implant. They subsequently tried to overcome the problem by putting a thin seed layer of titanium on top of metal one. When they etch through the oxide and reach the titanium, it shows up clearly as a white surface, and they know to stop. With this procedure, they have greatly increased their yield. In order to make a mask set to produce a double metal pad detector for MVD, three of the existing masks would have to be re-done. The cost of design and mask production would be $12,500, and three prototypes would cost $2500 each. I returned with an official price quote. My recommendation is to pursue evaluation of the two different pad designs, single and double metal, in a time frame that would allow us to choose between them for the production run.
We gave them a set of specifications for the MVD silicon strip cables. They had some questions concerning the tradeoff between our electromagnetic shielding requirements and our specs for intertrace capacitance. The better we shield the cable the higher the trace capacitance goes. We promised to firm up these specs. We also agreed to provide them with a precise definition of the overall thickness of the cable.
They have the capability to make the cables with either standard wire bond pads or with lead-frame windows. In lead-frame bonding, a rectangular window is created at the bond-end of the cable, exposing the traces and removing the kapton from below them. The lead frame is aligned over the bond pads of the detector or the IC to which it is to be bonded, and the bond is then made with a special wedge tool that can be fitted into a standard wire bonder. The trace is thus bonded directly to the pad. The advantage of the method is that for each bond, the tool performs one operation, whereas with standard wire bonding there are two operations, one for each of the pads the wire connects together. This advantage is graetest when the number of bonds is large, ie millions. It is not so clearly an advantage for the MVD. We asked them produce 2 or 3 cables for us with the traces terminating in standard wire bond pads. They estimated delivery of the cables by February of 1996. There will be no charge.
Runolfsson commented that a potential drawback for lead frame bonding to silicon strip detectors was the problem of aligning the cable to the detector before bonding. If one doesn't glue the cable first, the alignment may change, but if one does glue, one has to be extremely careful that none of the glue winds up on the lead frame. He agreed that standard wire bonds seem less problematic for the MVD applicaation. He also agreed to evaluate the bondability of the cables when they arrived at CERN from Kharkov. We discussed which glues, epoxies or silicones, would be best for our purpose. He offered to experiment with different glues for us as well.