Reprlnted from

NUCLEARINSTRUMENTS

&METHODSIN PHYSICSRESEARCH

8ectionA

Nuclear Instruments and Methods in Physics Research A 344 (994) 180-184North-Holland

A digitaI front-end and readout microsystem for calorimetry at LHC

C. Alippi ., G. Appelquist ;, S. Berglund ;, C. Bohm i, L. Breveglieri ., S. Brigati "P. Carlson h, P. Cattaneo f, L. Dadda ., J. David d, L. Del Buono d, A. DeU'Acqua "M. Engstrom i, G. Fumagalli " U. Gatti " J.F. Genat d, G. Goggi ',M. Hansen "*,H. Hentzell c, I. fIoglund k, S. Inkinen " A. Kerek h, H. Lebbolo d, O. LeDortz d,

B. Lofstedt " F, Maloberti " P. Nayman d, S.-T. Persson t, V. Piuri ., F. Salice ., M. Sami .,A. Savoy-Navarro d, R. Stefanetli ., R. Sundblad i, C. Svensson b, G. Torelli "J.P. Vanuxem " N. Yamdagni ;, J. Yuan b, R. Zitoun d

Il CERN, Geneua, Swirzerlondli Departmenl of Physia and Measuremenl Techno1qgy, Univers;ty of LinkopiJtg, Sweden" CenleT for Industrlal MicroeLectronics and Materlals Technology, University of Linkoping, Swedend LPNHE Universities Paris VI-ViI, Paris, Francee Dipartimento di Elettronica, Politecnico di Milano, ftaly, Sezione INFN, Pavia, Italyf Dipartimento di Fisica Nucleare e Teorica dell'UniuersiJa, Sezione INFN, Pavia, ltalyg Dipanimenlo di Elettronù:a deJl'Uniuersila e Sezione INFN, Pavia, /la/yh Manne Siegbahn lnstitute ofPhysics, Stockho1m. Swedozi Fysikum, Unioersity of Stockho!m, Swedenj SiCon AB, Linki5ping, Swedenk ABB.Hafo AB, liirfii/kl, Swet/enl Swedish MicroSysterns AB, Link6ping, Sweden

-,

Nuclear Instruments and Melhods in Physics Researcb A 344 (1994) 180-184North-Holland

NUCLEARINSTRUMENTS

&METHODSIN PHYSlCSRESEARCH

SecllonA

A digitaI front-end and readout microsystem for calorimetry at LHC

C. A1ippi c, G. Appelquist " S. Berglund " C. Bohm i, L. Breveglieri c, S. Brigati g,

P. Carlson b, P. Cattaneo " L. Dadda c, J. David d, L. Del Buono d, A. Dell'Acqua a,

M. Engstréim i, G. Fumagalli l, U. Gatti g, J.F. Genat d, G. Goggi l, M. Hansen a•• ,

H. Hentzell c, L Héiglund k, S. Inkinen " A. Kerek h, H. Lebbolo d, O. LeDortz d,

B. Lofstedt a, F. Maloberti g, P. Nayman d, S.-T. Persson l, V. Piuri c, F. Salice c, M. Sami c,

A. Savoy-Navarro d, R. Stefanelli c, R. Sundblad i, C. Svensson b, G. Torelli g

J.P. Vanuxem a, N. Yarndagni i, J. Yuan b, R. Zitoun d

Il CERN, Geneva, Swilzerlondb Deparlmenl of Physics and Measureme11l Technology, Uniuersity of Linkoping, Swedenc unte, for IndILstriai M"lCroeJecrronir.s and MOleria/s TecJmology, University of Linkòping, Swedend LPNHE Uniuersities Paris VI-VII, Paris, Francee Dipartimellto di EleUro'lica, Politecnico di Mi/ano, l(aly, Sezione INFN, Pavia, [taryf Dipartimento di Fisica Nucleare e Teorica d&'Uniuersito, Sezione INFN, Pavia, lta/yg Dipartimento di Elettronica deJJ'Uniuersila e Sezione lNFN, Pavia, lta/yIl Mantle Siegbalm Institute of Physics, Stockholm, Swedeni Fysikum, Uniuersity of Slockholm, Swedenj SjCon AB, Linkòping, Swedenk ABB-Hafo AB, farfalla, SwedenI Swedish MicroSys/ems AB, Linkiiping, Sweden

A digitaI solution 10 tbe front-end eIectronics for calorimetrie detectors al future supercolliders is presented. The solution isbased 00 bigh speed A/D converters, a fuUy programmable pipeline/digital filter cbaio and local intelligence. Questions of errorcorreetion, fault-Iolerance and system redundancy are also being considered. A system integralion of a multichannel device in amultiehip, SiJicon,oD-SilicoD Microsystem bybrid, is used. This solution allows a new level of integratioD of complex analogue anddigitaI functions, with an excellent flexibiliry in mixing tecbnologìes for tbe different funetional blocks. Il also allows a high degreeof programmability al both tbe fonetioo aod tbe system leveL, and offers tbe possibility of euslomising tbe mierosystem witbdeteetor-specific functions.

l. lntroduCtiOD

The electronics for signal deteetion and triggeringfor experiments at future proton super colliders will besome cf the most complex systems conceived so far forhigh energy physics. For calorimeters in particular,complexity, speed and decìsion (trigger) power require­ments cali for an exceptional level of data identifica­tiOIl~ compaction and processing already at detectorlevel. That, and the requiremems of massive triggerprocessing early in the acquisition chain, require thefront-end stage of these detectors to include a majorponion of the total eleetronics system.

* Corresponding author.

Front-end signa! processing for calorimetr}' detec­tors is essential in order to achieve adequate seleetivityin the trigger functions, with data identification andcornpaction before readout being required in the highrate environment of a high luminosity hadron machine.Other crucial considerations are the extremely widedynamic range and bandwidth requirements for tbefront-end electronics, as well as tbe volume of data tobe transferred to tbe following stages of tbe trigger andreadout system (1]. These requirements are best met byan early digitisation of the detector information, fol­lowed by on-chip digitai signal processing and buffer­ing functions al both the first-level and seccnd-Ieveldecision latencies.

In the following we will present FERMI [2], adedicated R &D project for digitai readout and pro·cessing of calorimeter output signals. ~

0168-9002/94/S07.00 <Cl 1994 - Elsevier Science B.V. AJl rights reservedSSDlOI68-9002(93JE1069-A

C. A/ippi et al. / Nud. blStr. and Meth. in Phys. Res. A 344 (]994) /80-184 181

2. Pbysics requirements and global parameters

tben Eq. (2) translates into a bound OD the derivativeof tbe transfer function,

.-' "

~

20001500

100lO

1000

1

500

0.1

,

,minimal I, -

~:0.03

,. mjnimar~ i I

..----.-- b : 0.1 , __----.---,---or: ' , I I ' , I , I ,1~ra~s~er,f~nc,li:n5.'1O

500

Vout(rnV)

2oo()~I 4-1

1500

1000

1000

E [GeV]

Fig. 2. Fractional resolution for Ibe FERMI bipalar compres­sor and ADe.

O.(l~

0.1

above the minimal transfer funetion, both in value andfirst derivative, and of having a monotonìc behaviour.

Practical implementations of nonlinear transferfunctions satisfying tbe bounds given by nrilin(E) can beobtained with piece-wise approximations having linearand/or logarithrnic segments. A practical compressoreircuit using a 4-fold piece-wise linear transfer funetioohas been implemented. Fig. 1 shows tbis transfer fune­tion, togetber with the "minimal" ones for two detec~

tor families:

A: a - 0.05 GeV, b ~ 0.10, c ~ 0.01;representative of scintillating fibre or liquid argoncalorimeters;

B: a - 0.02 GeV, b ~ 0.03, c - 0.005;representative of high performance scintillaling crystalcalorimeters.

Tbe "minimal" functions result trom integratingexpression (3b) in the energy range between 0.1 GeVand 2 TeV, wilh k ~ 1/li (corresponding to an upperIimil of R(E) - 0.08). All transter CUrves are nor­malised to two counts at E = 0.1 GeV.

v. READOUT ELECfRONICS

E (GeV)

Fig. l. Acrual and minimallransfer functions.

cr(E)/r,E -=:11

(5)

(2)

(1)

(3b)

(3a)

LSB(E) ,;;ud(E)/k.

dn(E)/dE ~ l/LSB(E),

dn( E)/dE" k/Ud(E).

lf n( E) is the (nonlinear) transter tunction ot thecircuit,

where a is tbe noise term, b a scaling tenn and c aconstane A quantisalion process matched to tbe detec­tor resolution implies:

un(E) -Ud(E)/k{i2, u,,,(E) ~ud(E) EIlun(E).(4)

Any practical implementation of a compression-con·verter system should produce a function R(E) boundedfrom above at alI vaJues of E.

The norrnalised integrai of expression (3b) can beused to define the minimal transfer function nmin(E)which provides the re((:vant bounds OD a wide family ofpossible transfer functions. Tbe detailed functionalfonn of any transfer function fulfilling tbese bounds islargely irrelevant. This holds tme provided it obeys thefollowing generai requirements at ali energies; of being

The combined effects of ali noise sources (includingconversion) OD tbe theoretical calorimeter performancecan then be described by tbe fractional increase of thedetector resolution:

The value of k is set by tbe allowed contribution of thequantisation noise Un' to the overall resolution:

The dynamic range for calorimetry in the multi-TeVregion is Iimited by noise (electronic and pile-up) ofabout 50 MeV, and the maximum energy deposit, whichis about J.5-2 TeV. This gives a dynamic range re­quirement OD the readout system of in the arder of 15lo 16 bits.

Il is not feasible, witb today's technology, to imple­ment AIO converters with more than lO bit resolutionat a sampling speed of 60-70 MHz as required by lhecol1ision frequency of the proposed super colliders.Hence, a compression of the dynamic range from 16 to10 bits must be done before applying the analoguesignal to the convener. This compression has to obeythe criteria outlined below.

The energy resolution u(E) of the detector is usu~

ally parametrised as a function of energy [3] as:

uiE) a b---=-ffi--+c

E E lE '

R(E) -u,o,(E)/ud(E)-1.

182 C. Alippi et al. / Nucl. Imtr. and Meth. in Phys. Res. A 344 (]994) 180-184

The quantization noise introduced by the compositetransfer function-digitisation described above is cam­pared to tbe detector fractional resolution for a type Aand B calorimeters in Fig. 2. OD tbe average, above 1GeV tbe intrinsic resolurion of the calorimeter in­creases by a relative factor in the order af a fewpereeot. With the appropriate choice of compressorparameters, similar results can be obtained for practi­cally aU calorimeter families. Il can be concluded framtbe above discussion that the combination of a suitablecompression scheme and a practical lO bit A/D con­verter G.e. with 9-9.5 effective bits) can be an ade­quate salution for early digitisation of the data frammost types of calorimetrie detcetors in !he LHC/SSCenergy range.

3. DescriplioD 01 !be FERMI strueture

The FERMI microsystem is designed to performdynamic rauge compression, 67 MHz digitisation, DSPand buffering af the data up to and including second­level trigger Latency. Il is intended to be rnounteddirectly OD the detector, along with a multi-FERMIcontralleI performing control, calibration and othersystem functions. Due to the harsh environment andlimited accessibility, the design will incorporate a highdegree of fault- and radiation-tolerance. Fig. 3 containsalI major FERMI components in the fcnn of a simpli­fied block diagram. It is divided iDto two maio parts:aD acquisition part and a comman service part. The

former consists of nine identical channels in theeegroups of theee channels, whiJe the latteI is unique.The acquisition channel begins with a nonlinear ampli­tier to achieve signa) compression. The compressedsignal is applied to the A/D converter wbere it issampled at the rìsing èdge of the internaI 67 MHzdock. This event also starts tbe digitisatiOD processoThe resulting lQ-bit data are linearized and expandedto the fuH 16-bit dynamic raoge linear representationby means of a 1024 word Look-Up Table containingthe inverse transfer function of tbe total detector- andfront end electronics system. Tbe expanded data arestored in tbe data memory at locations supplied byexternal circuitry CJx>inter from tbe address generator).

External circuitry keeps track of wbere aH samplesare stored until the first level trigger has made itsdecision. If a sample point is accepted as a valid event,a predefined temporal environment (a time frame) iskept in the memory while 10catioDs that do Dot belongto 5uch intervals are released. Released locations arerecycled to be used for new ioput storage.

Extemal circuitry is also responsible for translatingthe event identifier in a read command into a pointersequence corresponding to the time frame in questionoThe pointer sequence, together with readout controlinfonnation, is transferred to the attached FERMIswhere it is stored in the Readout controller. It isenvisaged that the reduced readout should be defaultwhen pushing data to the second level trigger proces­sorso These could, however, returo far selective read­out (pull) of full time frames, which might be necessaryin order to resolve more complex (pile-up) conditions.

Leve! 1Triggef

Anak>gi,",,",

Anak>ginpu1

Address

Wrile ReadPointersaodControl

Level2andLeve! 3

FERMI

. /

Fig. 3. An averview cf FERMI.

------------------------~----------_._-

C. AUppi et al. / Nucl. [nstr. and Meth. in Phys. Res. A 344 (1994) 180-184 183

Fig. 4. Block diagram of tbe first level trigger filter.

3 5no 01 subfilrers

Fig. 6. Filter companson.

crlElfE [Xlr-

0.00\ ~1

e

y

tim

The data transfer from FERMI can tbus be charac­terised as "push with selective pull".

4. Filte...

Tbe filter functions of FERMI consist of two differ­eol 61teT sections and of memory elements lo stofe thedat3_ while waiting for the trigger decisions. Tbe fusllevel trigger fiJter (Fig. 4) is designed lo recognise asignal giving precise timing informatioo and relativelycoarse energy informatioo. lt operates OD the suro ofthe nine channels and uses two five-lap finite impulseresponse (FIR) filters together wìth a maximum findeI.The second level trìgge! filter is designed lo give aprecise estimation of the pulse amplitude in conditionswhere the sample timing jitter is the dominant sourceof errar. Its output is used for the second level triggerprocesso The bank of tbree !inear FIR filters and arank operator are used. A block diagram of the secondlevel trigger filter is shown in Fig. 5. It is optimisedeither by using an analytical method or by numericaliteration. The performance as a function of the Dum­ber of banks is shown in Fig. 6.

5. Faull tolerance

The fault tolerance in FERMI is implemented indifferent ways depending OD tbe logical function ofeacb stage. It can be automatico as in memories witherroI correction code (ECC), or it may require con­troller Ìntetvention. In the latteI case, on-line CITar

checking will alarm the controller, which will respond

with a hardware reconfiguration, thus eliminating thesource af error.

For crucial parts, redundant hardware, like triplica.tioo of latches, is used. In arithmetic parts, such as thccircuits for first level summation aud digitai filters,results will be checked by applying modulo-3 errorcbecking.

6. Statos or tbe proje<t

A bipolar noolinear compressor amplifier prOlo­type, completely fulfilling aH specifications, is designed,implemented and tested. A CMOS version is designedand hardware is expected during summer 1993.

A two-stage pipelined A/D converter has beendesigned and implernented, and is tested at sub-partlevel.

Two prototype versions of an parallel successiveapproximation A/D converter have been designed andimplemented; ODe of the versions, completely fulfillingali specifications, has been tested at the sub-part leveland as a complete converter while the other has beenpartly tested while a complete prototype is currentbeing processed. Delivery is in May 1993.

Tbe digitaI p.n is modelled using rbe VHDL lan­guage and is being verified usiog XILINX pro­gramrnable gate arrays as test media.

The filter sections have been defined and are veri­fied by simulation. VHDL models are being wriuen.

Test and calibration procedures are being workedaut, p.rtly as resulls of tbe VlIDL modelling, p.rtly .sresults of continuously ongoing system-wide simula­tions.

A floor-plan for a complete microsystem is beiogdesigned.

Fig. 5. The second level filter.

7. Conclusions /summary

In this paper wc bave demonstrated the feasibilityof producing calorimeter readout electrooics, based 00

early digitisation, for tbe next generation of hadroncolliders. Thc digitai represeotation of the data allowsthe generation of very sharp thresholds for the triggerprocesses. The problerns of handling signals with very

v. READOUT ELEcrRONlCS-~

184 C. Alippi et al. / Nucl. [mlr. and Meth. in Phys. Res. A 344 (1994) 180-184

high dynamic range· have been solved by the use of anonlinear transfer function matched to the resolutionof the particular detector. A/D converters fulfillingthe requirements, special1y thase which concern speed,size and power consumption, are being developed.Digitai circuirs providing the necessary storage, pro­cessing and communication functions are descnbed.The questions of fault tolerance have been addressedal each stage of tbc design. A silicon-on-silicon mi­crosystem packaging technology [4] is used lO achieve acompact, ye! flexible, rcariaut system.

References

[l} V.G. Goggi and B. Lofstedt, Proc. ECFA Large HadrooCollider Workshop, CERN 90-10, ECFA 90-133, voI. 3(1990) p. 190.

[21 The FERMI collaboratioo, CERN/DRDC 92-96 (992).[3] For a review see: R. Wigmans. Annu. Rev. Nucl. Parto Sci.

4i (1991) 33.[4] A Novak et al.. A standard micr<h-ystem module, Pre·

sented at the ISHIM Meeting, Nordic Chapter, À1and,Finland, September 1990.