CALICE MAPS Meeting, RAL, 10/10/05 ================================== Present: Jamie Crooks, Paul Dauncey, Renato Turchetta, Mike Tyndel, Giulio Villani, Nigel Watson Minutes: Paul Minutes of previous meeting: No corrections. The action items had been covered except the work on simulation combining Giulio, Nigel and John had not yet started. Sensor simulation: Giulio had circulated an outline of the sensor simulation plan since the last meeting; this is linked from the CALICE-UK main page: http://www.hep.ph.ic.ac.uk/calice/maps/ChargeCollectionSimulationsProp.doc He also showed the results of a simulation of a 150x150mu^2 array of 6x6 diodes spaced 25mu apart. By summing in analogue the signals on four diodes to make an effective 50x50mu^2 pixel, he improved the signal size by around a factor of two. The neighbouring pixels show signals around 30% of the hit pixel, even though the MIP was deposited close to the edge of the 50x50mu^2 pixel. The diodes simulated were 1.5x1.5mu^2. Making these bigger might improve the charge collection slightly but would increase the capacitance and hence noise. There is a significant amount of work to be done to study and optimise the diode layout for maximal signal/noise without losing too much charge and generating too much heat. The time for 90% of the charge to diffuse out was found to be ~160ns. Giulio had not grounded the substrate in his simulation; in this case, it should be tied by the p-wells to ~200-300mV above the epitaxial layer potential. If this works, it is very convenient for the mechanical construction. Physics simulation: Nigel showed some plots from his student, linked from the meeting web page. The basic result is that they find 25x25mu^2 pixels appear to have significantly better linearity than 50x50mu^2, let alone 100x100mu^2; see pages 16 and 19. The main items noted were: o There is no cross talk in the simulation. This will clearly change the results, particularly for small cell sizes. The simulation should include input from Giulio's work (above). It was not clear to what extent the cross talk for inclined tracks is caused by the epitaxial thickness or the charge diffusion; Giulio's work implies the latter is a significant effect. o The energy resolution is not the most important measure for the ECAL design. The hadronic jet resolution is not limited by the ECAL resolution and the better quantity to study is the PFLOW ability. However, the software to study this is not yet available. o In a dense shower where the non-linearity effects will be most significant, there is probably a high likelihood that neighbouring pixels will also be hit. A study could be done to see the average number of charged tracks crossing a pixel as a function of the number of the eight nearest neighbour pixels above threshold. It is possible this could lead to a non-linearity correction at a local pixel level; this would be robust against any distortions a global correction might have on PFLOW. Giulio will subdivide the ionisation position into 5x5mu^2 subareas and output the average charge sharing for each of the 25x25mu^2 pixels. This can then be used by Nigel to parametrise the cross talk. Only a triangle of the subareas needs to be simulated by Giulio due to the symmetry of the diode arrangement. It was decided that to make progress, we should fix some parameters for now and go round a loop of studying the resulting performance. Following this, we will want to iterate the parameters. The chosen ones were an epitaxial layer of 20mu and a diode spacing of 25mu, which can be added later to give 50x50mu^2, 75x75mu^2 and 100x100mu^2 effectivel pixels after the simulation. The substrate (i.e. wafer) thickness should be fixed to 330mu to match the existing wafers in the SID detector simulation. The sensor area should be 2x2cm^2 with a 500mu dead edge strip along two sides (see below), giving a 5% dead area. As specified in the requirements, up to 5% additional dead area can be simulated through randomly distributed dead pixels. Requirements: Jamie showed some slides resulting from the discussions he and Renato have had following the circulations of the requirements; these are linked from the meetings web page. Achieving 10% dead area might be tight. A band of 500mu on two sides would be preferable to 250mu around all four sides for the common circuitry but this would still be very tight for space. The proposed scheme for storing timestamps has the memory location counter local to each pixel, so each can independently store different numbers of timestamps. A maximum from 4 to 16 timestamps is foreseen. This requires the location counter and pointer logic to be implemented in each pixel, which could take a large area. Jamie suggested an alternative where the location logic is done one time globally for a large number of pixels. This would require the train to be divided up into (e.g.) eight periods in time and the logic would advance to point to the next memory location after each 1/8 of the train. This would mean any pixel with more than one hit during the 1/8 of the train would lose data. How this compares with the local logic requires some study; Paul will do a simple Poisson calculation to compare the two. Paul showed some (old) plots from a talk giving an overview of the basic idea of the mechanical arrangement for the diode pad option: http://www.hep.ph.ic.ac.uk/calice/others/011031daresbury/011031.ppt There have been some changes in detail since this time but the general concept is similar. Some specific comments on the questions raised by Jamie: o The data rates are clearly a major driver and we will need some estimate of what dominates the hit rate, i.e. physics or noise. o Paul will check the analogue readout ASIC power specifications with the French groups this week when he is at DESY. However, we should not consider the power as a critical constraint for the first round of sensor design; however, it should obviously always be a consideration. o The issue of how long we need to store the data within the pixel drives the issue of DRAM vs SRAM. Actions: o Giulio and Nigel to do 5x5mu^2 parametrisation and integrate into the simulation. o Paul to feed back info to Renato and Jamie on data flow rates and Poission probabilities. Next meeting: 13:00 on Mon 7 Nov at RAL. Meet at Renato's office (R76, first floor) at 12.00 for lunch. Mike will be unable to attend. [Changed after meeting to 10:00 on Thu 10 Nov at RAL]