Cost effective electronics for lar and photo-detectors readout

The T600 ICARUS detector has a DAQ system that has proved a quite satisfactory performance in the test run performed in Pavia in summer 2001.

The electronics has been described in various papers and technical notes. In this paper, starting from the experience gained in the T600 operation, we propose an upgraded DAQ scheme that implements the same basic architecture with more performing new components and different modularity in view a multi-kton TPC (e.g. MODULAr) with a number of channels in the order of ~n_*10⁵.

Also the electronics for PMTs detecting scintillation light in Ar will be shortly presented.

1 T600 DAQ

The ICARUS T600 read-out architecture was designed for a detector with few 10⁴ channels.

The analog chain consists of a front-end low noise charge sensitive pre-amplifier, based on a custom designed BiCMOS dual channel IC with j-Fet input stage, followed by a baseline restorer to reduce low frequency noise. The gain of the front-end amplifier and filter is 1V/164fC. The 10bit ADC input range is 1V, and then the least count is equivalent to 1000 electrons. This value matches with the amplifier noise of ~1250 electrons with a detector capacitance of 450pF (signal wires plus cables).

The basic modularity of the system is 32. The digitizing stage consists in a set of 32 channels connected to four 8-to-1 multiplexer, each one followed by a 10bit ADC sampling at 20MHz. The sampling time is then 400ns per channel.

For each group of two ADCs the two 10bit outputs are merged in a 20bit word. As the positive clock edge drives one group of two ADCs and the negative clock edge drives the other group of two, 20bit words can be transmitted at 40Mhz rate on single 20 bit wide channel. This scheme is implemented on a single VME-like analog board (CAEN V789) hosting the 32 channels amplifiers, ADCs together with a 20bit,

40MHz, serial link that allows sending the data to a second digital board (CAEN V791) that performs buffering of events. The serial link is LVDS (Low Voltage Differential Signaling) and serializes the 20bit data and clock on four copper lines with an equivalent data throughput of 800Mbps. The block diagram is in Fig. 1.

The receiving digital board, named ARIANNA, hosts two custom-made feature extraction ASIC chips (DAEDALUS) for hit finding, zero skipping and self-triggering. Each of the two DAEDALUS chips operates on the data stream of a set of 16 channels and controls also the circular memory buffer. DAEDALUS includes also a median filter to reduce high frequency noise. A 28bit absolute time register is associated to each data buffer in the memory to allow alignment of data in the event reconstruction. The board complies with the VME electrical and mechanical standard.

This architecture has proven to be well suited for the T600 detector and to perform accordingly the expectations.

2. 
   Critical issues
   

It is important to remember that the DAQ architecture for T600 was conceived in 1997 and DAEDALUS and front-end dual channel BiCMOS circuits were designed in 1998. The full system (~50,000 channels) was designed and built from 1999 to 2000.

Even if the architecture proved to be very performing, nowadays it would not be possible to replicate the same circuits, as they are, because many basic components are now discontinued.

In this scenario it is very important then to make a critical analysis of the whole system to spot the areas where the necessary changes could led to a more efficient structure.

Adopting a top down approach one evident limitation of the T600 DAQ is due to the adoption, perfectly legitimate at that time, of the VME standard for the digital boards. For the analog boards the same 6U Eurocard standard was adopted, with a custom backplane to connect the inputs from wires and distribute common signals (ADC baseline bias, enable signals, test pulses, etc.).

In the T600 DAQ 18 ARIANNA digital boards are housed in one VME crate that serves a total of 576 channels. One crate is connected to an analog crate, with the same modularity, which in turn receives the signals from a single T600 flange that in fact has 18 feed-through connectors, one for each set of 32 channels. Configuration and control of the 18 boards relies on a dedicated VME cpu, which also handles the data transfers from board buffers to the Ethernet event builder network. Performance of the DAQ system is bounded by the VME (8-10 MB/s), together with the sequential order single board access mode inherent to the shared bus architecture.

The redesign of the system can overcome this limitation adopting a modern switched I/O, as PCI Express standard for instance, allowing the parallelization of the data flows. In addition such I/O transaction can be carried over low cost optical gigabit/s serial links. This allows a more effective modularity of the digital hardware architecture, decoupled from the geographical distribution of the signal feed-through, thus lending to a larger integration and the consequent lower cost per channel.


FIGURE 2 here
Fig. 2: Input amplifier equivalent circuit

3 The amplifier
Considering now the analog front-end does it not seem that a completely new design could improve either the S/N or the density (channels per package).

In a multi-kton TPC we can foresee wires with a pitch larger than the 3mm used in the T600. The adoption of 6mm pitch for MODULAr seems reasonable and will permit to use most of the existing molds and tools for wires support.

The capacitance associated to each channel will be determined by the capacitance of the wires, in the order of 20pf/m, in parallel with the capacitance of the cable, in the order 50pF/m. Let's assume in the following discussion ~600pF as a reasonable value for 10m electrode wires, 6mm pitch, and average 8m of cable.

The dominant noise in a high capacitance detector, [5] [6] [7] [8], is the series noise e_sn (voltage noise) linearly increasing with the input total capacitance (C_D) while the parallel noise (current noise) contribution is proportional to the shaping time of the signal.

The equivalent circuit showing the effect of the input capacitance on the serial noise is given in Fig.2.

The serial noise generator has a noise density

where g_m is the transconductance of the input device. At the amplifier output the noise N becomes

The best component would be the bipolar transistor that typically exhibits transconductances in the order of

Second in the list would be j-Fet transistor that however has much lower transconductance, in the order of

However the bipolar transistor has a typical parallel noise density i_p= 1-2pA/Hz^.5 that with signals with a time extension of the order of 6-10µs, like in a liquid Argon TPC working at 500-1000V/cm, gives an unsustainable noise contribution.

The j-Fet then, with its quite reasonable transconductance and negligible parallel noise (i_p=.001 pA/Hz^.5, three order of magnitude lower), is an obliged choice. In terms of serial noise density the typical figure for a suitable component is ~1-2 nV/Hz^.5.

Normally we don't even consider CMOS for the extremely low transconductance and unfavorable merit factor. However recently, for space application, CMOS have been used for low temperature design. In this workshop an interesting poster on a front-end amplifier for use in liquid Ar has been presented [n]. The issue of mounting electronics inside the dewar will be presented later.

In liquid Argon a m.i.p. releases ~4000 electrons per mm, that means 2 fC/wire in case of 3mm or about twice for 6mm pitch. The Icarus amplifier, designed with dual j-Fet input, has an ENC of ~1250 electrons at C_D=450pF.

In Fig.3 an image from a LAr TPC is shown as an example. In Fig.4 signals along adjacent wires, belonging to the short track in the black box, are given. The overall gain at the ADC output is

~6 counts/fC, being the r.m.s. noise 1.2 counts (~1250 electrons at C_D=450pF) we reach a S/N=~10. With a pitch of 6mm we could keep the same S/N.

The custom IC [8] designed for the T600 front-end electronics in BiCMOS technology, has the typical structure of an unfolded cascode integrator, allowing however the connection of external j-Fet at the input. In principle it would have been possible to include the j-Fet inside the monolithic device but unfortunately in BiCMOS technology j-Fet are obtained as a byproduct and their characteristics are totally unsatisfactory in term of gain and noise not to mention the fact that the silicon surface needed for a j-Fet with the required transconductance would be of the order of ~0.5mm².

In the T600 electronics we connected two discrete j-Fet in parallel to reach a total g_m of the order of 50-60mS. This solution seems suitable also for large TPC with C_D less than ~1000pF.

According to (3) advantages of cold electronics are the reduction of the input capacitance because of shorter cables, the increase of the j-Fet transconductace g_m, and the reduction of the temperature T. In Figure N the typical transconductace curve as a function of the temperature is given. The improvement of the S/N will be a factor 2.4 due the combined effect of lower Johnson noise [~1.9] and 60% higher transconductance [~1.26].

However the effect of shorter cables will be limited as it is not conceivable to mount each amplifier at the end of the corresponding wire. Even inside the detector the amplifiers will be grouped on cards, so the only part of connection saved is the one from the detector feed-through to the external electronics crates. A proper design of electronics crate for external electronics would eliminate external cables.

A real disadvantage of cold electronics is the limited design architectures due the limited choice of active components working at 86ºK, largely compensating the improvement of S/N.

But the reason why the cold electronics will not pay is another. A multi-kton TPC will require tens of years for R&D and construction and eventually will run for other tens of years. It is not conceivable to prevent any possible upgrade of the only part of the detector that could be modified and could benefit of the technological evolution of electronics in the future decades.

A new front-end layout for a future large detector can be hosted in a compact crate very close to the feed-through flange. Or even one could think to use the flange itself as a backplane supporting the analog boards. This solution is under study and a possible layout will be shown. The number of connectors and cables would be drastically reduced with a benefit for cost and S/N.

The new modularity will be the number of channels served by one feed-through flange. In the solution under development we use the modularity of the Icarus flange that serves 576 channels.

This new 576 channels module under study must also perform the data conversion before transmitting data to the digital buffering board as will be described.

These devices are quite interesting for price, power consumption and dimensions. Typically they are packed in Mini Small Outline Package (MSOP) smaller than 5x5 mm².

The choice is rather large and we can expect that more products will be available within one year.

The acceptable sampling frequency for a TPC with 6 mm pitch can be assumed in the range of 1-1.5Mhz for which there is already a wide choice of devices.

Assuming a sampling frequency of 1.5Mhz, and 10bit ADCs, and data compression in one byte, we need to transmit 6912Mbit/s, that on an optical fiber, to balance zeros and ones becomes 8640Mbit/s. Optical links with 1.5Gbit/s data rates are standards and can be driven by the RocketIO™ interfaces available on many FPGA from different vendors. Six optical links could serve all the channels of one module (576) and convey also extra information as absolute time. Some of the links will be bidirectional to distribute absolute clock and simple commands.

The ADCs and FPGAs will be housed in the same crate next to the flange or even on special boards on top of the amplifiers boards shown in Fig.6.

What will not be discussed in this paper is the implementation of the equivalent of the ARIANNA board. One could say that nowadays for hit finding ASIC VLSI will not be required as the feature extraction algorithm can be implemented in powerful FPGA.

5. 
   Photo-detectors readout
   

Ionization in liquid Argon (LAr)is accompanied by scintillation light emission. The two processes are complementary and their relative weight depends on the strength of the electric field (recombination) and dE/dx.

Electron yield, Y_ion=~2.9*10⁴ e^-/MeV, and photon yield, Y_ph=~2.4*10⁴ γ/MeV, are similar for a m.i.p. at an electric field of 500V/cm.

Light is emitted at 128nm, its detection generally requires wave-length shifters, with two components exponential decay: one short time constant, τ_s ~ 6ns, and one long time constant, τ_L ~ 1.5μs.

Prompt light signal is typically used for trigger (e.g. Icarus) or for particle identification (e.g. WArP). In fact for a m.i.p. about 25% of emitted light has short time constant and about 75% has long time constant, while a recoil has about 75% of the emitted light with short time constant and about 25% with long time constant. In Fig. 7 integrated signals for electron like events and ion recoils are given.
FIGURE 7 here
Fig. 7: Integrated signals for electron like events and ion recoils.

In double phase (liquid and gaseous Argon) detectors, one can distinguish the primary scintillation, S1, in liquid, from the scintillation in gas, S2, proportional to extracted ionization electrons. S1 and S2 have similar characteristics and can be studied in detail identifying individual electrons through high frequency, high priced, waveform recorders with sampling frequency in the gigahertz range. In Fig. 8 typical S1 signals (left) and S2 signals (right) from WArP detector are shown.

We can expect in the future the appearance on the market of competitive products at reduced price.

The layout foreseen for charge signals that uses the flange as electronics backplane can be exploited also in case of photons signals.

[1] Carpanese C. et al., ARIANNA: the ICARUS Experiment Readout Module. IEEE Trans. on Nucl. Science. vol. Vol. 45 no. 4 (1998), pp. 1804-1808 ISSN: 0018-9499.