The APV25 is a low-noise, radiation-hard front-end readout chip
made in a 0.25um CMOS process, designed for beam-synchronous
operation at 40MHz in the CMS Experiment at LHC. Each of its
128 input channels features a 192-cell analog pipeline as well
as a switched capacitor filter where the weighted sum of three
consecutive samples is calculated.
Using this deconvolution circuit, the preamp/shaper transfer function
is essentially undone in a numeric way, which leads to just a single
non-zero output value (25ns wide) compared to the shaper output which
is approximately 160ns wide. Hence, it is possible to unambiguously
identify the bunch crossing from which a particle originated.
If the APV25 clock and the bunch crossings are not synchronous,
e.g. in case of a quasi-continuous beam, this method fails.
However, the APV25 can also be operated in a multi-peak mode where
still three consecutive samples are stored in the pipeline with
each trigger, but the deconvolution filter is turned off and all three
samples are read out. By applying several triggers spaced by 75ns
(minimum trigger distance), up to ten triplets (=30 consecutive
samples) can be read out, which represent the shaper output waveform.
In order to reconstruct peak time and amplitude, a fit can be applied
to each event. This method was successfully demonstrated with APV25 data
obtained in beam tests and resulted in an RMS time resolution of
approximately 2ns at a cluster signal-to-noise ratio of 25.
The innermost layer of the BELLE Silicon Vertex Detector at KEK
(Tsukuba, JP) currently suffers from high occupancy in the order
of 10%, which stems from the fact that its VA1TA readout chip has
a shaping time of about 800ns. In summer 2007, an upgrade is
planned for the two inner layers where the silicon sensors essentially
remain the same, but the readout will be done by APV25 chips.
Its faster shaper already reduces the occupancy by a factor of 12.5
(not exactly the ratio of shaping times due to deviations from the
ideal CR-RC shape). By using the proposed peak time finding method,
comparing the hit timing to the trigger time and discarding off-time
hits, another factor of up to 8 (depending on the S/N ratio) can be
gained. Hence, the projected occupancy will be well below 1% in the
current environment, allowing headroom for future luminosity increase.
Obviously, a fit applied to each hit is not suitable in an experiment.
Hence, we developed lookup tables which fulfill the same purpose,
perform almost equal to the fit function, but much faster, and are
very easy to implement in FPGAs.
Finally, we will also present the VME-based data processing boards
which are called "FADCs" but actually do much more than just
digitization. Several Altera FPGAs perform channel re-ordering,
pedestal subtraction, a 2-pass common mode correction, hit finding
(zero suppression) and time finding on the APV25 data. All functions
are pipelined, such that the modules are dead-time free. Since the
signals are sparsified on the module, the output data rate is
considerably reduced compared to transparent readout.