The APV25 Readout Chip for the CMS Strip Tracker

Imperial played a leading role in the RD20 collaboration, which investigated silicon tracking sensors, readout electronics and radiation hardness of possible technologies. Our initial activities were investigations of radiation tolerant microstrip sensors where work in RD20 laid foundations on which many other groups have built. It became evident that a bigger issue would be development of radiation-hard front-end electronics where we, partly through long-standing close links to engineers in Rutherford Appleton Laboratory, had a lot to offer. RD20 devised the deconvolution method, which was an Imperial innovation, and prototyped CMOS circuits which showed it could provide practical LHC readout. We worked closely with RAL Instrumentation Department on design and evaluation of the series of circuits which culminated in the front-end APV25 readout chip. Imperial took particular responsibility for analogue parts of the design, particularly amplifier and shaper, and all aspects of evaluation, including all radiation tolerance studies, most of which were very original and at leading edge of activities in the field. To give one example, single event upset studies were the first observations of this phenomenon on a real tracker readout circuit, and the first really relevant to LHC experiments.

The road to the APV25 was long, with Imperial performing prototype evaluation of Harris, DMILL and, finally, IBM 0.25 µm CMOS processes and detailed investigations of each of them. There was much scepticism at most stages, especially of the risk and challenge in embarking on the 0.25µm CMOS route. In the event it proved to be remarkably smooth and with substantial benefits to CMS, in cost, quality, performance and power. The pioneering work by a strong Imperial-RAL-CERN team was productive and significant in convincing many others in the HEP community to follow, and the IBM technology has been almost universally adopted by CMS, in particular in a major revision of the ECAL system begun in 2002.

Imperial was able to steer CMS towards 0.25µm CMOS because we had a leading role in the Tracker readout development from the beginning, based on two R&D projects in which we participated: RD20 and RD23, on optical links. The major elements of the readout and control system were ASICs, optical links and a digital readout board (Front End Driver), which was developed by a team of Imperial physicists and RAL engineers, with several students. This was also an ambitious project, which exploited the rapid growth of programmable digital (FPGA) technology in recent years, and culminated in the board now in large scale production.

Following R&D and prototyping, manufacture took place during 2003 and 2004; all APV25 chips were tested at Imperial using an automated probe station, and completed in 2005. Over 150,000 die from ~650 wafers have been tested, ultimately at a rate of 2 wafers per day, providing ~120,000 Known Good Die. Although only ~75,000 chips are required by CMS, assembly losses were significant because of sensor and hybrid problems. During 2002 and 2003, yield variations on some APV25 lots were a concern and intensive studies were carried out, in collaboration with IBM. This concluded by identifying a process refinement to optimise a crucial Chemical Metal Polishing step and metal-metal contacts were uniform and reliable. After this, the typical production yield has been ~90%, which is remarkable compared to initial expectations. A detailed report has been given in a NIM paper, as well as several intermediate publications in LHC Electronics Workshops. The large number and high quality of APV25 chips allowed us to respond to requests for its use in other projects. Initially US government export restrictions on the CERN contract related to the radiation tolerance made this difficult but in the last year or so this has relaxed considerably. COMPASS (CERN) and Belle (Japan) have each been provided with ~1500 die, while smaller numbers are in use in ZEUS, STAR, and TOTEM.

The front-end APV25 readout chip.

The APV25 chip has 128 channels, each containing a preamplifier and shaper with a 50ns peaking time, followed by a 192 location memory into which samples are written at 40MHz. Locations of data awaiting readout are flagged so they are not overwritten. Following a trigger, three samples from the memory are processed with the APSP deconvolution filter which re-filters the data with a shorter time constant. This confines the silicon signal to a single beam crossing interval and allows to measure signal amplitude and assign the event time. The chip can be operated in three modes: peak mode, in which the output sample corresponds to the peak amplitude following a trigger, deconvolution mode, in which the output corresponds to the peak amplitude of the APSP filter, or multi-mode where, following a trigger, three samples are read out. After the APSP data are held in a further memory buffer prior to switching through an output analogue multiplexer. This is required so that one event can be multiplexed out while another is prepared for transmission. The APV also contains system features including programmable on-chip analogue bias networks, a remotely controllable internal test pulse generation system and a slow control communication interface.

The chip has been extensively tested in small-scale laboratory set-ups, several beam tests at various beam facilities and, now that the experiment has reached the construction phase, on a much larger scale in numerous integration centres. The first beam test using an LHC-like 25 ns bunched beam was performed in 2000, which was essential in validating the performance of components and close-to-final assemblies. The beam tests continued productively over several years, identifying many details which might have degraded the system. Among them was the discovery of the effect of heavily ionising knock-on particles in the silicon sensors, which, because of very large pulse heights, can momentarily saturate the amplifiers and give rise to unwanted common-mode effects. Careful evaluation in beam tests and the Imperial laboratory proved that the impact on CMS will be very small and well within tolerable limits.

Page content provided by R. Bainbridge.