CALICE MAPS Meeting, RAL, 20/10/06
==================================
Present: Jamie Crooks, Paul Dauncey, Anne-Marie Magnan, Konstantin
 Stefanov, Renato Turchetta, Mike Tyndel, Giulio Villani, Nigel Watson,
 John Wilson

Minutes: Paul


Minutes of last meeting: No corrections.
 The visit of people from Foundry B to RAL was postponed to Nov.
 The new RAL/PPD member, Marcel Stanitzki will start in early Nov.
 Konstantin has posted some plots (originally from Chris Damerall)
 showing the probability distributions for the liberated charge in
 silicon of various thicknesses; see usual web page.


Sensor design: Renato has been given a price of $80k to develop and
 qualify a deep p-well process with Foundry B. This does not include VAT
 but this may not be applicable as the company is outside the EU and
 will not deliver anything to RAL. The ownership of the rights of the
 resulting process was not clear; this will need to be clarified but
 it should certainly grant RAL the ability to use the process for all
 of its MAPS projects. The development would take around one month
 although it is not clear if it could be done in time for the Jan run
 or if it would be technically possible to use it with other designs
 on the same wafer. Renato has already asked Foundry B about the Jan
 run but has had no reply yet.

 The cost is still quite high compared with our budget (but smaller than
 the $200k quoted by Foundry D). It would require us to use our working
 allowance which needs approval from our Oversight Committee. The next
 OsC meeting is on 28 Nov. As this process would be beneficial for other
 MAPS designs, then RAL may be prepared to pay part; this would
 certainly make getting OsC approval easier. Renato will look into this
 and Paul will contact the OsC before the next meeting to inform them of
 the issues.

 The deep p-well process needed would be a complementary version of an
 existing, qualified process for deep n-wells. In this, an n+ layer is
 put down below the whole of an n-mos transisitor so that the p+ area
 is completely cut off from the epitaxial layer. The n+ layer thickness
 is around 2mu, which therefore reduces the epitaxial layer thickness
 in the area below the transistor, reducing the signal charge by ~15%.
 However, given that signal gains of a factor of two would be expected,
 then this is clearly acceptable.

 It was agreed that if we are going to use deep p-well for the second
 sensor fabrication round, then we should also use it for the first.
 This means we observe the realistic signal size as soon as possible and
 don't have to extrapolate the performance so far. This puts constraints
 on the fabrication schedule (see below).

 Jamie (after Giulio below) showed some slides of the simulations he had
 run to study the circuit performance; see usual web page. The two
 designs were as discussed in the previous meeting; design 1 is Renato's
 "PreSample" circuit while design 2 is Jamie's "PreShape" circuit. For
 the PreSample circuit, the noise is found to be independent of the
 input capacitance but the signal is not. In contrast, for the PreShape
 circuit, the noise is a strong function of the input capacitance but
 the signal seen is not. The noise quoted for design 2 in the previous
 meeting of 6mV was found to be incorrect and a value of ~11mV is now
 expected, which gives a S/N of less than 5 with the present parameters.
 The n-well size is much larger than assumed in the sensor simulations
 for design 2 so without using deep p-well, this would reduce the signal
 even further.

 Design 1 needs an explicit reset after each hit seen although charge
 below the threshold could build up, shifting the effective threshold
 level, before a hit is observed. For design 2, the shaper preamp will
 step up with each charge deposit during the spill. A MIP gives ~10mV
 step but the range is ~1V which allows a lot of overhead.

 Mike raised the point that the event rate at the ILC is low, so the
 signal does not have to be as fast as 150ns. The noise will go down
 as the square-root of the shaper speed in design 2 so a factor of four
 in the time constants for the shaper to give a signal over ~600ns will
 give a factor of two improvement in the S/N. Although it would be
 desirable to keep a fast shaper, it is not a critical parameter and so
 should be lengthened to get a usable S/N.


Sensor simulation: Giulio has used the results produced by Jamie to look
 at expected S/N values; his talk is on the usual web page. He assumes a
 parasitic 8fF capacitance and so, for a 1.8x1.8mu2 diode area, finds a
 noise of 57 ENC. This compares with signal sizes in the worst case of
 around 250 electrons.

 He has been assuming that only charge accumulated within 150ns could be
 used, but there should be no problem with using a later sample (300ns
 or even 450ns) after the event beam crossing as this is just a constant
 offset. The pile-up this causes will be present in any case as it is
 caused by the physical diffusion times. However, as the physics event
 rate at the ILC will be low, pile-up should not be an important issue.

 Giulio has a simulation for 3.6x3.6mu2 currently running and will do
 the same analysis for this. Following this, he proposes to go to the
 eight-diode layout detailed in his talk. However, to speed up the
 simulation, he should reduce the number of points and the area of the
 detector simulated. Explicitly, 7 (rather than 21) points were thought
 to be sufficient, namely the centre point and then three points equally 
 spaced from the centre to the middle of one edge and also three points
 equally spaced from the centre to a corner. In addition, for these
 studies, we do not need to know the details of where charge lost to the
 pixel goes, so he can simulate a single 50x50mu2 pixel, rather than
 nine pixels. However, from now on, he should do each simulation both
 with and without the central n-well. Despite this last doubling, the
 other reductions should make each simulation substantially quicker.

 Jamie will calculate the equivalent input capacitances for each of the
 geometries Giulio simulates, so that a S/N value can be extracted for
 each.

 Giulio will also reanalyse his existing data to see how the charge
 collection looks when the n-well is in the corner rather than the
 centre. However, given that the lowest charge is collected for a hit
 near the edge, rather than under the n-well, then this geometry may
 just give a very inefficient region.


Physics simulation: Nigel showed some plots from Yoshi on hits in 48
 contiguous pixels, here contiguous in z, i.e. parallel to the beam;
 see the usual web page. A figure for the average (non-zero) energy in
 a pixel of 7.2keV was given, which is substantially above the 3.7keV
 most probable value for a MIP at normal incidence. This could be due
 to speeds off the MIP value or angled tracks but could also be due to
 the long high-end tail of the Landua for a thin layer. The mean value
 seen in 48 contiguous pixels is 9.2keV, showing that normally only one
 pixel is hit of the 48.


Valencia ECFA meeting: Nigel also showed an outline of Yoshi's talk plan
 for the Valencia meeting which starts on Mon 6 Nov; see talk and link
 on the usual web page. The talk concentrates on the simulation
 verification and does not contain any charge diffusion or PFA
 modelling results. Comments will be fed back to Yoshi by Nigel.

 In addition to this talk, which will be in a simulation session, it
 would be good to start publicising the MAPS work more in the ILC
 community. A more technical MAPS talk in the calorimetry session at
 Valencia would be ideal; something at the level of Giulio's recent
 Siena talk would be fine. Giulio himself is not able to go to the
 Valencia meeting. Konstantin is already going; although he is
 relatively new to the project, it was agreed that he would be able
 to give a well-informed talk and hence should do this if he was willing
 to do so.


Schedule: If we are aiming for the Jan shuttle run deadline, then layout
 would have to start at the beginning of Nov and even this is tight.
 This means there is no time for further design; all that could be done
 is to choose one of the two designs, tidy up the loose ends and then
 start the layout. However, it is not yet known if the deep p-well would
 even be available for this shuttle run and there are several items
 which are still be to investigated, so this sounds very unlikely.
 Renato will let us know by email when he finds out about the deep
 p-well timing but it seems a working assumption should be that we have
 to now go for the Apr run. This means the IDR will now not happen on
 Tue 31 Oct but will be rescheduled for around the end of Dec or early
 Jan.


Next meeting: This will be either Mon 13 Nov or Tue 14 Nov, depending
 on which is prefered by Mike and Renato (who were not present at the
 end of the meeting).

 [Note added after the meeting: the date was fixed for Mon 13 Nov.]