mailSend Mail to Chuck Britton

Multiplicity-Vertex Detector Electronics Development for Heavy-Ion Detectors

C. L. Britton, Jr., W. L. Bryan, M.S. Emery, M.N. Ericson, M.S. Musrock, M.L. Simpson, J.W. Walker, A.L. Wintenberg, F. Plasil, G.R. Young1
Oak Ridge National Laboratory

M.D. Allen, L.G. Clonts, E.J. Kennedy, R.S. Smith
The University of Tennessee

J. Boissevain, B.V. Jacak, J. Kapustinsky, J. Simon-Gillo, J.P. Sullivan,
H. van Hecke, N. Xu
Los Alamos National Laboratory

Published In: Proceedings of the First Workshop on Electronics for LHC Experiments, Lisbon, Portugal, September 11-15, 1995, pp. 84-87.
PHENIX-MVD-96-12

ABSTRACT

This paper presents the electronics work performed to date for the Mulitplicity-Vertex Detector (MVD) for the PHENIX collaboration at RHIC. The detector consists of approximately 34,000 channels of both silicon strips and silicon pads. The per-channel signal processing chain consists of a preamplifier-gain stage, a current-mode summed multiplicity discriminator, a 64-deep analog memory (simultaneous read-write), an analog correlator, and a 10-bit 5s ADC. The system controller or Heap Manager, supplies all timing control, data buffering, and data formatting for a single 256-channel multi-chip module (MCM). Each chip set is partitioned into 32-channel sets. Prototype performance for the various blocks will be presented as well as the ionizing radiation damage performance of the 1.2 nwell CMOS process used for fabrication.

I. INTRODUCTION

The Multiplicity-Vertex Detector (MVD) of the PHENIX detector at RHIC is one of the most challenging of the PHENIX detector subsystems. The requirements of low-power consumption, small physical area, a channel count of approximately 34,000, and flexible data handling have resulted in a challenging set of problems. The requirements of the system are to have signal-to-noise not less than 20:1 for a 1MIP signal which results in noise less than 2500 electrons rms. The added requirement of being able to discriminate on a 0.5MIP even for the per-channel multiplicity discriminator is much more difficult.

The vertex position is located using hits in the barrel for Z position between -40 and +40 cm, which covers the 2 length of the interaction region. The vertex position resolution is better than 200, limited by the 200 pitch of the silicon strips. This is more than adequate for the muon and central tracking use.

The multiplicity trigger is formed using the sum of silicon channels (both pads and strips) above a threshold corresponding to approximately 0.25-0.33 of a minimum ionizing particle (MIP). For off-line analysis of the charged particle multiplicity and fluctuations over 5 units of rapidity, the pulse height from each channel will be used to separate multiple from single hits. In a single event, the MVD is sensitive to multiplicity fluctuations at the 10% level; this is limited by backgrounds and charged particle statistics rather than MVD performance.

This paper presents measurements of the prototype component systems of the MVD detector electronics. The circuits include the preamplifier-discriminator, analog memory unit (AMU), analog-digital converter (ADC), and heap manager.

II. ELECTRONICS ARCHITECTURE

A block diagram of the electronics is shown in Fig. 1. This includes the preamplifier, discriminator, current-sum output, analog memory-correlator, and ADC. The overall controller, or Heap Manager, is not shown but will also be discussed.

Figure 1. System block diagram.

The electronics will be mounted on multi-chip modules (MCM). Each MCM will be connected to 256 detectors and will contain 8 preamplifier-discriminator chips and 8 analog memory-ADC chips. Each of these chips will contain 32 channels of their respective functions. In addition, the MCM will contain the Heap Manager chips and associated control logic. The present prototype chips are 8 channels of respective functions.

A. Preamplifier

The preamplifier, versions of which are presently being used by not only PHENIX, but PHOBOS (also at RHIC), and the Naval Research Laboratories (NRL) for Germanium spectroscopy [1], is presented in Fig. 2. It utilizes a PMOS cascode amplification stage for low 1/f noise. The input device operates in a weak inverson with a drain current of 100A. The drain current curve is shown in Fig. 3. The PHENIX and PHOBOS versions employ wideband gain stages after the preamplifier because the output will be processed by a correlated sampler. The NRL version employs a semi-Gaussian shaper with 7s peaking time. A summary of the performace of the three versions is presented in Table 1. The resultant power dissipation of the PHENIX preamplifier with the gain stages is approximately 1.2mW. The circuit is fabricated in 1.2 n-well CMOS and has a pitch of 85.

Figure 2. MVD preamplifier

Figure 3. Drain current vs. Vgate-source

ParameterPHENIX (224ns double correlated)PHOBOS (Wideband)NRL (7s peaking)
0pF noise660e750e205e
Slope69e/pF30e/pF3e/pF
Gain44mV/fC4.2mV/fC11mV/fC
Drain100A200A100A
current

Table 1. Performance of various preamplifer versions.

B. Discriminator

The multiplicity discriminator block diagram is presented in Fig. 4. Each time an input step greater than the threshold setting is detected, a fixed-value current source is switched to a summing node. The current-mode outputs of 256 discriminators are summed in this manner to produce a multiplicity output whose amplitude is proportional to the number of fired discriminators.

Figure 4. Block diagram of the discriminator

The summed currents are input to an on-MCM opamp that converts the current to a voltage which is then run off of the MCM to a pipelined flash ADC for conversion of the trigger sum. The design allows triggering at every beam crossing (112ns intervals) with no deadtime. The total power consumption is approximately 500W/channel. A plot of the output is shown in Fig. 5.

Figure 5. Discriminator output

C. Analog memory-correlator

The analog memory unit (AMU) is a 64-cell deep, voltage-write-voltage-read, deadtime-less topology with a power dissipation of approximately 1mW/channel. The memory is followed by an analog correlator that performs the double-correlated sampling. Both direct memory output and correlated output are available to the subsequent analog-digital converter (ADC). The memory reading and writing is addressed externally from the Heap Manager. The read and write logic decoders are completely independent to allow simultaneous read-write (deadtime-less) operation. A block diagram of the memory is presented in Fig. 6.

Figure 6. Memory block diagram

D. Analog-digital converter

The readout architecture of the subsystems within the PHENIX detector have been intentionally designed to resemble each other a great deal. This has been done for two reasons. The first is for maintainability. Using the same parts in many subsystems reduces the needed stocks of spares and simplifies troubleshooting. The second is cost. We can take advantages of economies of scale by fabricating a large quantity of one part and use it in many subsystems within the detector. This philosophy is best exhibited in the AMU/ADC chip. We are designing a universal AMU/ADC chip that will be used in several subsystems. The architecture of the chip is 32 channels of 64-deep AMU and 32 channels of 12-bit ADC. The ADC will be switchable for 10-, 11-, or 12-bit operation. For the MVD, the ADC will be set for 10-bit operation for dynamic range requirements.

The analog-digital converter (ADC), presented in Fig. 7, is a 10-bit Wilkinson type [2] with a maximum dynamic range of 4.5 Volts. This prototype has eight channels.

Figure 7. Block diagram of ADC chip.

The Gray code counter uses a positive ECL logic level differential clock input with frequencies as high as 212 Mhz. A Gray code counter circuit is used for two reasons: first, since only one bit changes at a time, transient currents should be smaller, consequently making noise levels lower. Second, any single-bit error in transmission will result in only an LSB magnitude error. The Gray-coded data passes through a Grapy-to-binary conversion circuit when the chip data is read out. There is also a one-word deep memory on the chip to allow readout of data from one conversion while another conversion is taking place.

The ramp generation circuit consists of an operational amplifier with an on-chip feedback capacitor to form an integrator. This provides a low output impedance ramp signal that is not affected by loading from multiple ADC channels on the same chip. The ramp and comparator circuits use a separate analog power and ground connection from the rest of the digital sections of the chip for isolation.

The ADC uses 3.75W/MHz/channel in the digital circuitry and 6.5mW total in the ramp generator. This results in a power dissipation of approximately 0.6mW/channel for a 32-channel chip or 1.19mW/channel for the 8-channel prototype at 100MHz clock frequency (5s conversion time for 10-bit conversions). At 144 Mhz clock rate and 10 bits, the Integral Nonlinearity (INL) is about +/-0.1% of full scale. Full Width Half Max (FWHM) is less than 2 bins wide at higher input voltages and deteriorates to about 6 bins at low input voltages. The cause of this has not been determined at this time.

E. Heap Manager

The Heap Manager controls all on-board functions including AMU address list management and control (for asynchronous read and write operations), correlator timing, Level-1 buffering associated with trigger latency, ADC precharge, acquisition and conversion control, data handling including data collection, formatting, and transmission, and command interpretation, execution, and timing. This function has been implemented using commercially available FPGAs allowing in-circuit programmability for future functional upgrades. A multi-function serial interface is used for FPGA programming, calibration setup, diagnostics, and monitoring.

A generic heap manager has been developed for use in all PHENIX detector subsystems using analog memory for front-end signal buffering during the Level-1 trigger decision. However, due to the reduced size and power requirements for the MVD, this functional block has been partitioned such that the minimum required electronics are physically located near the detectors. The remaining circuitry resides at the ends of the detector subassembly as interface cards to the data collection modules (DCMs). Figure 8 is a block diagram of the MVD heap manager showing the basic functional blocks and system partitioning.

Figure 8. Block Diagram of the Heap Manager

III. RADIATION EFFECTS

Some significant (>20krad) radiation exposure of MVD is expected due to the proximity of the detector to the interaction region. Because of the desire to utilize a standard non-radiation hardened process, measurements of radiation effects upon the target 1.2 CMOS process were made. Versions of the preamplifier, analog memory and ADC circuits have been irradiated to various doses of ionizing radiation. The process appears to be sufficient for doses up to greater than 75krad for most functions. The major problem is in the leakage current of the input protection for the preamplifier. Devices used for protection networks begin to exhibit increased leakage current due to the radiation-induced decrease in the nmos threshold voltage. We are continuing to investigate improved input protection. Figure 9 is a plot of the analog memory unit Integral Non-Linearity (INL) for various doses of 60Co.

Figure 9. Amu integral-nonlinearity for various radiation doses.

IV. CONCLUSIONS

The circuits for the PHENIX MVD has been presented. The readout consists of preamplifier, multiplicity discriminator, analog memory, ADC, and Heap Manager controller. Prototypes of the major functional blocks have been fabricated and have been tested. The radiation tolerance of the CMOS process is presently under test.

VI. ACKNOWLEDGMENTS

The authors would like to thank Norma Hensley for her help in preparing this manuscript.

VII. REFERENCES

  1. R.A. Kroeger, W.N. Johnson, R.L. Kinzer, J.D. Kurfess, S. Inderhees, M.D. Allen, G.T. Alley, C.L Britton, L.C. Clonts, M.N. Ericson, and M.L. Simpson, "Charge Sensitive Preamplifier and Pulse Shaper Using CMOS Process for Germanium Spectroscopy," IEEE Trans. Nucl. Sci., Vol. 42, No. 4, Aug. 1995 (921-924).
  2. A.L. Wintenberg, T.C. Awes, C.L. Britton, Jr., M.S. Emery, M.N. Ericson, F. Plasin, M.L. Simpson, J.W. Walker, and G.R. Young, "Monolithic Circuits for the WA98 Lead Glass Calorimeter." Presented at the 1994 Nuclear Science Symposium, Norfolk, VA, Nov. 1994.

Research sponsored by the U.S. Department of Energy and performed at Oak Ridge National Laboratory, managed by Martin Marietta Energy Systems, Inc. for the U.S. Department of Energy under Contract No. DE-AC05-84OR21400.