CALICE MAPS Meeting, RAL, 30/11/06
==================================
Present: Jamie Crooks, Paul Dauncey, Anne-Marie Magnan, Yoshi Mikami,
 Owen Miller, Marcel Stanitzki, Konstantin Stefanov, Mike Tyndel, Giulio
 Villani, Nigel Watson

Minutes: Paul


Minutes of last meeting: No corrections.


Foundry news: The foundry has sent a lot of detailed process
 information to RAL, which will allow Giulio to do a more precise sensor
 simulation. The epitaxial layer doping is less than assumed, resulting
 in higher mobility and so faster charge collection. The p-wells are
 thicker (1.7mu) than previously assumed which will reduce the thickness
 of the epitaxial layer for charge creation by a small amount.

 The foundry have also started working on the deep p-well process
 already, on the expectation that we will definitely sign the contract.
 The actual cost will be $100k but will not require VAT to be paid, so
 this is cheaper than the worst case we previously considered. The cost
 to CALICE should be around 20k, with RAL/EID paying the other 35k. We
 have the approval of the OsC for this now so the contract should be
 signed as soon as possible. The main issue is now the exclusivity; it
 seems we will have the process details on Jan 5 and so can design to
 them from that date. All other customers will get this information only
 six months later. There are also issues of patenting which need to be
 settled. The foundry will require us to use some (the exact number is
 not yet known) of our process splits to test the deep p-well process
 parameters.

 The epitaxial layer can be chosen from several standard values (3.3mu,
 5mu, 12mu or 20mu) even for a MWP run. The 5mu value is their standard
 but we will use 12mu. If possible, we would want to try the 20mu as
 one of our process splits, if we have enough. It might even be worth
 buying more wafers to allow a larger number of process splits than the
 standard six to allow this; Jamie thought the incremental cost would
 be low.


Sensor design: Jamie showed some slides of recent work, see the usual
 web page. He finds either circuit (albeit for different reasons) will
 require considering hits over several time bins, probably up to 1us.
 Whether this is acceptable is an issue for the simulation, although
 it is likely to be OK given the low event rate and noise rate we are
 aiming for. The shift in time is dependent on the signal size, so the
 time offsets in principle give information but it is not clear how
 useful this would be.

 Jamie finds potential mismatches of components in the comparator mean
 the effective threshold could vary with a width of ~20%. This could
 produce large differences in noise and/or efficiency so he proposes
 a 4-bit trim on the threshold voltage for every pixel. The total
 amount of configuration data needed to set these is 4 bits per pixel
 x 10^12 pixels = 500GBytes and so is large but not unfeasible. However,
 the bigger issue is how to calibrate the trim setting. Ideally, they
 would be set to give equal efficiency on every pixel. If the noise is
 directly proportional to the gain (as is usually the case) then this
 would mean that equalising the noise rate would be the method to use.
 This can be done quite quickly as even at a 10^-6 noise rate, then 
 to get ~100 noise hits per pixel (and so give a reasonable estimate of
 the rate) needs ~10^8 bunch crossings, which is ~10^4 bunch trains,
 taking less than one hour per threshold setting at a bunch train rate
 of 10Hz. Obviously, higher noise rates would be even quicker. However,
 if the noise is not a good estimate of the gain, then the calibration
 becomes much harder.

 Jamie proposes a monostable to give a fixed output pulse length after
 a hit which resets the circuit, removing the need for this to be done
 elsewhere. This draws significant current but only when switching so
 the total power it takes is dependent on the hit rate and so hard to
 estimate accurately.

 Jamie and Marcel have drawn up a preliminary list of the sensor I/O
 signals but only for the second sensor round. Paul asked for such a
 list for the first sensor (for development of the DAQ system) a while
 ago so Jamie will send that round soon. This will be defined by the
 time of the IDR.


Interim Design Review: Both of the suggested external reviewers from the
 Microelectronics group are not available on the IDR date of Mon 18 Dec.
 We had previously ruled out Tue 19 Dec and the people present could not
 make Fri 15 Dec, so Jamie will find which of Thu 14 Dec or Mon 18 Dec
 is possible for getting two external reviewers and fix the date. This
 should be done asap.


Sensor simulation: Giulio has run the simulation with a 3.6x3.6mu^2
 diode size and finds a worst case MIP signal of 400e-, again near the
 pixel edge. Jamie estimates the noise for this diode size to be 33e- in
 the shaper circuit and 40e- in the sampler. This configuration is then
 the best in terms of S/N.

 The next step is to upgrade the simulation with the new information
 from the foundry to make it more realistic. Giulio will then rerun the
 same geometry as above to see if there are significant differences. He
 will then go to bigger diode sizes, to see where the S/N starts to fall
 off. The suggested next step was another factor of two in the diode
 side length, i.e. 7.2x7.2mu^2.


Sensor testing: During the OsC meeting on Tue 28 Nov, Phil Allport had
 suggested using a monoenergetic gamma source rather than cosmics to
 give an absolute energy calibration. This had previously been dismissed
 due the lack of an external time indicator, i.e. a hit in a
 scintillator, which meant effectively integrating over all samples.
 Phil stated that LCFI do this and get an acceptable signal. Konstantin
 reported that LCFI use Fe55, which has a pair of gamma lines at 5.9 and
 6.2 MeV. Gammas of this energy interact within a few mu in silicon so
 all the charge would be deposited in the circuit or the first few mu of
 the epitaxial layer. This would then not accurately mimic the charge
 deposited by a MIP. This needs further thought.

 Paul showed a few slides from Matt Noy; see usual web page. Matt 
 suggests using a simple FPGA board being developed at Imperial for
 other projects. (Note, this is not the I-DAS board previously
 considered, as there were insufficient of these available for our
 needs.) This board should be cheap (around 250 pounds) compared with
 commercial FPGA development boards with similar amounts of I/O. One
 limitation of the board suggested was the lack of on-board memory. If
 this cannot be added straightforwardly, then solutions would be to use
 the internal FPGA RAM or to put a memory onto the "generic" interface
 board which will be needed between the suggested board and the PCB
 holding the MAPS sensor. The memory and speed requirements are modest
 so either solution would be possible but to avoid limitations, then
 Paul will check on the possibilities of installing a significant amount
 of memory on-board.

 [Note added after the meeting: Jamie circulated a spreadsheet he uses
 to calculate the memory storage of the sensor; see usual web page. He
 estimates a total on-sensor memory of 60kBytes for the first sensor and
 500kBytes for the second.]


Physics simulation: Yoshi showed some slides of GEANT4-based
 resolutions; see usual web page. By counting the number of pixels with
 physical energy deposits of any size, he finds an "ideal" energy
 resolution of around 23%/sqrt(E). The figures suggested in the usual
 calorimeter studies are more like 15%/sqrt(E) but these tend to use a
 fitted Gaussian, ignoring tails, while Yoshi used a more conservative
 RMS. Note, the data at 500MeV may be incorrect as a particle of that
 momentum would not be expected to reach the ECAL in a 4T field.

 Anne-Marie showed some plots of resolutions vs threshold from the
 digitisation-level simulation; see usual web page. She uses the charge
 diffusion results from Giulio's simulations to simulation charge in
 neighbouring pixels. However, Giulio simulates the charge being
 deposited along a 32mu vertical line with 80e-/mu, giving a total of
 2560e-. Of this, most of the charge in the bulk below the epitaxial
 layer is lost, while some charge from the first few mu of the bulk 
 gets into the epitaxial layer and so can be seen as signal. Anne-Marie
 needs the total amount of charge able to diffuse in the epitaxial layer
 so as to normalise Giulio's numbers and this is not given directly by
 the simulation. She uses 1300e-, which was considered a reasonable
 estimate. GEANT4 works in terms of physical energy deposits with a
 MPV for MIPS of around 3.2keV. Hence, one e- is equivalent to
 approximately 2.5eV, so her assumed noise of 120eV is roughly 50e-,
 which is a little higher than Jamie estimates. One suprising result is
 that the smearing due to diffusion completely dominates over the noise
 when considering resolution. This implies a low noise will mainly be
 useful for reducing the data volume.

 Nigel showed some slides of work he is doing with Owen on machine
 backgrounds; see usual web page. The e+e- background rates are very
 high but all the pairs produced are at low momentum and low angles to
 the beam so very little reaches the calorimeter. So far, only 0.1 of a
 bunch crossing has been simulated but none of the particles produced
 would give a hit. Clearly, some pre-GEANT4 cut will be needed to allow
 a large number of particles to be quickly simulated while only then
 studying in detail the rare cases which might result in calorimeter
 hits.


Conferences: Yoshi will give a MAPS presentation at the silicon
 detectors for calorimetry workshop in Seoul, Dec 14-16. He will give a
 practise talk at Birmingham on Mon 11 Dec in the afternoon; a phone
 will be available for people who want to listen in from outside. The
 talk is 16mins plus questions. The talk should be a general overview
 and not concentrate on just simulation; hence an updated version of
 Konstantin's Valencia talk would be appropriate.

 The LCWS07 meeting at DESY from 30 May to 5 Jun will include an ILC
 "review" of calorimetry R&D. The review has no official power to
 approve or reject projects but this is an opportunity for us to further
 advertise the MAPS work. However, there are significant issues about
 how we present this with regard to the other CALICE ECAL work, as we
 have to be seen to be part of CALICE. Paul has already been considering
 this and will continue the dicussion with the CALICE Steering Board. It
 might also be an opportunity to submit an LC-Note on MAPS; if so, then
 this should include:
  o Description of the sensors (electronics, dead areas, etc)
  o Optimisation of geometry (pixel size, diode size, diode placement)
  o System outline (size, mechanics, power, etc)
  o Physics and noise rates and DAQ requirements (backgrounds, intrinsic
    noise, etc)
  o Physics performance (PFA and EM resolution)
 All of this work is already planned to be done although it will be
 tight to complete it on this timescale.


AOB: The funding for WP3 was specified in detail when the grant was
 submitted; there is around 1k for setting up each of the test benches
 for each round of sensor testing. However, the total amount in the WP
 is fixed. Besides the sensors themselves, the next largest item is the
 DAQ systems. Hence, if this can be done below the budgetted cost (hence
 the desire to use the cheap Imperial DAQ board) then the saving could
 be used for other items. However, we need to know the DAQ costs before
 any funds could be released. Giulio stated that the laser system
 probably does not need any further expense although he could not rule
 out some small items in the future; these would clearly come out of the
 1k line item. Note, although the grant will cover the DAQ PCs, it does
 not cover desktop computing; this is an institute cost. Central MC
 production or data analysis can be done either at the RAL computer
 centre or on the Grid.


Next meeting: The IDR will be either the 14 or 18 Dec. The next normal
 meeting will be Fri 19 Jan.
 [Note added after the meeting: the IDR was fixed for Tue 19 Dec.]