Particle Detectors IV: Silicon Micro-strips and Calorimeters

A modern detector is ordered by how much damage each layer does. Whatever measures without disturbing has to come first and whatever measures by destroying has to come last, and that fixes the architecture.

🎯 Why this matters

It is also why upgrades are painful. Nothing on the inside can be reached without removing everything around it, so the innermost layer is the one an experiment has to get right before it starts.

Two more detectors and the chapter’s instrument catalogue is complete. They are the two that a modern experiment is mostly made of, and they are opposites: the silicon micro-strip measures a position without disturbing the particle, and the calorimeter measures an energy by destroying it.

Silicon micro-strip detectors

Take the drift chamber and replace two metres of gas with a hundred micrometres of silicon. The micro-strip detector , developed in the 1980s, is a wafer 100 μm or so thick and a few square centimetres in area, carrying a ladder of many $n$ – $p$ diodes laid down as parallel strips at a pitch of tens of micrometres. Reverse-bias it until it is fully depleted, and a charged particle crossing it makes electron–hole pairs that drift to the strips, where charge amplifiers read them out.

The strips are, structurally, the anode wires of an MWPC. Everything else about the comparison is different, and worth being precise about.

⚙️ Engineer’s bridge — silicon trades gain for statistics

A gas chamber has a gain of $10^5$ and needs it: a minimum-ionising particle makes only about 97 ion pairs per centimetre of argon, which is a 10 % counting fluctuation before the avalanche adds its own. Silicon has no gain at all, and does not want any:

	argon at n.t.p., 1 cm	silicon, 100 μm
energy per pair, $W$	26 eV	3.6 eV
density	1.66 mg/cm³	2.33 g/cm³
pairs created	≈ 97	≈ 10 800
$\sqrt N/N$	10 %	1 %
gain needed	$10^5$ , with avalanche fluctuations on top	none
collected charge	1.6 pC (after gain)	1.7 fC (raw)

Silicon makes a hundred times more carriers in a thousandth of the thickness, because its pair-creation energy is seven times smaller and it is 1400 times denser. The signal is small in coulombs — femtocoulombs, so you need a genuinely quiet charge amplifier — but it is large in carriers, and carriers are what statistics counts.

This is the same trade an engineer makes choosing between an avalanche photodiode and a low-noise PIN diode: gain is only worth having when the primary signal is too small to survive the amplifier’s noise floor. In silicon it is not.

Where it breaks: gain becomes worth having again as soon as the primary signal shrinks. Thin sensors, or timing measurements needing a fast rising edge, have pushed silicon back towards internal multiplication — the low-gain avalanche detectors now going into the HL-LHC timing layers are exactly that. And silicon carries a failure mode a photomultiplier does not: radiation damage. The bulk is progressively destroyed by the fluence it is measuring, so the innermost layers are consumable items with a replacement schedule, which is not a trade-off any amplifier-noise argument would have predicted.

The spatial resolution is about 10 μm. Note that this beats what a strip pitch alone would give — 25 μm of pitch quantised as $d/\sqrt{12}$ is 7.2 μm, 50 μm gives 14.4 μm — because the charge from an inclined track is shared between neighbouring strips and the ratio interpolates between them. The digital resolution of §1.13c is the worst case; analogue read-out beats it.

What 10 micrometres is for

📐 Physics you need first — why a lifetime becomes a distance

An unstable particle produced with momentum $p$ travels, on average, $\ell = \beta\gamma\, c\tau = (p/m)\,c\tau$ before it decays (§1.6–1.7). So a lifetime — something measured in seconds — turns into a length you can see, provided your detector’s resolution is smaller than it.

Charmed and beauty particles have lifetimes of a fraction of a picosecond, which sounds hopeless: $10^{-12}$ s is not a time anyone measures directly for a single particle. But $c\tau$ for a $D^0$ is 123 μm, and at a few GeV the boost stretches that to most of a millimetre.

That is the entire argument for the vertex detector. You are not measuring a time. You are measuring a displacement, and converting.

🔢 Worked example — how many sigma is a charm decay?

Take a $D^0,\ \bar D^0$ at $p = 10$ GeV. Its mass is 1.865 GeV, so $\beta\gamma = 10/1.865 = 5.36$ , and with $c\tau = 123$ μm:

\ell = \beta\gamma\,c\tau = 5.36 \times 123\ \mu\text{m} = 660\ \mu\text{m}.

Against a 10 μm resolution that is a 66 σ separation between the production and decay vertices. A $B^0,\ \bar B^0$ at the same momentum gives 860 μm, or 86 σ.

This is why the book calls the device essential rather than useful. With a 1 mm resolution the measurement is impossible; with 10 μm it is not even difficult. Nothing changed about the physics — only the ruler.

To do that you need four or five micro-strip layers, so that the two vertices can each be reconstructed as the intersection of several tracks rather than guessed. This stack — the vertex detector — goes immediately after the target in a fixed-target experiment, and wrapped around the interaction point in a collider. It is what made the top quark (§4.10) and the physics of the $B$ mesons (§8.6) accessible.

Calorimeters

A calorimeter measures the energy of a particle or of a group of particles, and the measurement is destructive: all of the energy must be released inside the detector, or you have not measured it. There are two kinds, electromagnetic and hadronic, and they differ only in which interaction drives the cascade.

Electromagnetic calorimeters

Electrons, positrons and high-energy photons shower (§1.11) through two processes that feed each other:

e^\pm + N \to e^\pm + N + \gamma \tag{1.122}

\gamma + N \to e^+ + e^- + N . \tag{1.123}

Bremsstrahlung turns an electron into an electron plus a photon; pair production turns that photon into two more electrons. The mean distance between events is one radiation length $X_0$ , and that single number governs everything.

The measuring principle is beautifully indirect: the total length of all the charged tracks in the shower is proportional to the incident energy. Track length is proportional to ionisation charge, and charge — or something proportional to it, such as scintillation light — is what you actually collect.

🌳 The cascade, and where 1/√E comes from

incident energy E₀10 GeV

material          lead:  X₀ = 5.6 mm,  E_c ≈ 600/Z = 7.32 MeV
E₀/E_c                       = 1367
depth of the maximum  t_max  = log₂(E₀/E_c) = 10.4 X₀ = 58 mm
particles at maximum  N_max  = E₀/E_c       = 1367
total charged track length   ≈ X₀·E₀/E_c    = 7.65 m
Molière radius  r_M = X₀·21.2 MeV/E_c       = 16.2 mm
resolution (Eq. 1.124)  σ/E  = 15–18 %/√E   = 4.74–5.69 %

Multiply the energy by ten and the shower gets ten times denser but only 3.3 radiation lengths deeper — depth grows as log E, which is the reason a calorimeter that works at 1 GeV still works at 1 TeV with barely any more material. The book's 15–25 X₀ covers that whole range.

particles in the cascade, N = 2^t

Only the multiplying phase is drawn — beyond t_max the particles are below E_c, stop doubling and die away over a few more X₀. Note the axis: the count is exponential in depth, so the DEPTH you must buy is only logarithmic in energy. That is the whole economics of calorimetry.

Model: one bremsstrahlung or pair conversion per radiation length, so the particle count doubles and the energy per particle halves. Multiplication stops when the particles drop below the critical energy E_c ≈ 600 MeV/Z and start losing energy to ionisation instead.

hadronic calorimeter, 40–60 %/√E
EM calorimeter, 15–18 %/√E
tracker of §1.13c, δp/p = 0.87 % × p

The two instruments fail in opposite directions. A tracker measures curvature, so its error grows linearly with momentum; a calorimeter counts shower particles, so its error falls as 1/√E. Below the crossing you trust the tracker, above it the calorimeter — and every experiment is built to put that crossing where its physics lives. The hadronic band sits a factor 2.7–3 worse than the electromagnetic one, because a hadronic shower loses an unmeasured and fluctuating fraction of its energy to nuclear binding and to neutrinos.

💡 What this really says — why the resolution improves with energy — and why nothing else does

Every other instrument in this chapter gets worse as the energy rises. The tracker’s sagitta shrinks as $1/p$ ; time-of-flight separation shrinks; Cherenkov angles saturate. The calorimeter is the exception, and the reason is that it is a counter.

A shower deposits its energy as $N$ detected particles, and $N \propto E$ . Counting $N$ things has a fluctuation $\sqrt N$ , so

\frac{\sigma(E)}{E} = \frac{\sqrt N}{N} = \frac{1}{\sqrt N} \propto \frac{1}{\sqrt E}.

That is not a property of any particular technology; it is Poisson statistics, the same $1/\sqrt N$ that governs photon shot noise, Monte Carlo error and every average you have ever taken. It is also why the destructive nature of the measurement is a feature: you must absorb everything, because anything that escapes is a particle you failed to count.

\frac{\htmlClass{t-sig}{\sigma(E)}}{\htmlClass{t-E}{E}} = \frac{\htmlClass{t-a}{15\% - 18\%}}{\sqrt{\htmlClass{t-E}{E}\,(\text{GeV})}}

(1.124)

At 1 GeV this is a 17 % measurement, at 100 GeV a 1.7 % one, at 1 TeV 0.54 %. Read it as: the calorimeter is the instrument for high energies, and hopeless at low ones.

Every symbol, one at a time

Hover or tap a symbol above — it lights up in the equation and its meaning, units and type appear here.

🔢 Worked example — read Fig. 1.28 as a measurement

The book’s Fig. 1.28 is a cloud-chamber photograph of an electromagnetic shower, with the chamber divided by eight lead plates 12.7 mm thick. The lead does essentially all the absorbing: $X_0 = 5.6$ mm for lead against tens of metres for the gas. So the total depth is

\frac{8 \times 12.7\ \text{mm}}{5.6\ \text{mm}} = \frac{101.6}{5.6} = 18\ X_0,

comfortably inside the 15–25 $X_0$ a calorimeter needs to absorb a shower completely. And you can read the initiating particle straight off the picture: there are no tracks before the first plate, so the incoming particle was neutral — a photon, which converted in the lead.

That is a calorimeter, built in 1952, that also happens to show you the shower. Modern ones show you a number instead.

📏 Sampling versus homogeneous — the one design decision

Sampling . Alternate plates of a dense absorber (lead, typically 1 mm) with plates of an active medium (plastic scintillator, several mm). Only the energy deposited in the active layers is seen; the majority, deposited in the lead, is never measured — you rely on the sampled fraction being proportional to the total. Cheap, compact, and the sampling fluctuation is what limits Eq. (1.124).

Homogeneous . Make the whole thing active: an array of prisms of a transparent dense medium such as lead glass, each long enough to contain a full shower, each read out by a photomultiplier at its exit face. The signal is Cherenkov light from the shower’s charged particles (§1.13a), proportional to the incident energy. Sampling fluctuations are eliminated, so the resolution improves — at the cost of an expensive detector-grade crystal for every cubic centimetre of your calorimeter.

CMS (§9.14) chose homogeneous: lead tungstate crystals, because the Higgs → γγ measurement lives or dies on electromagnetic energy resolution.

Transverse size, and a number that barely changes

Depth is set by $X_0$ ; width is set by the Molière radius, defined as the radius of the cylinder containing 90 % of the shower’s energy:

\htmlClass{t-rm}{r_M} = \htmlClass{t-x0}{X_0}\left(\frac{21.2\ \text{MeV}}{\htmlClass{t-ec}{E_c}}\right)

High density gives narrow showers — which is what you want when two photons must be told apart.

Every symbol, one at a time

Hover or tap a symbol above — it lights up in the equation and its meaning, units and type appear here.

💡 What this really says — everything is about two centimetres wide

Put lead into the formula: $X_0 = 5.6$ mm and $E_c \approx 600/82 = 7.3$ MeV give $r_M = 16$ mm. Now iron: $X_0 = 17.6$ mm and $E_c \approx 23$ MeV give $r_M = 16$ mm again.

That is not a coincidence. $X_0$ shrinks with $Z$ and $E_c$ shrinks with $Z$ in almost the same proportion, so the ratio hardly moves: the Molière radius of essentially every dense material is 1.5–2 cm. The book quotes 21 mm for the lead tungstate crystals of CMS, which are built with diameters of that order for exactly this reason.

So one number, roughly the same everywhere in the periodic table, fixes the transverse granularity of every electromagnetic calorimeter that has ever been built. If you want finer position resolution than 2 cm, you get it by measuring the shape of the deposit across several cells, not by making the cells smaller.

Hadronic calorimeters

Hadronic calorimeters measure the energy of a hadron, or of a jet — the narrow spray of hadrons a quark turns into (Chapter 6). Since the jet carries essentially the quark’s energy, this is the instrument that measures quarks.

The principle is the same as the electromagnetic case with one substitution: the cascade is driven by strong interactions, so the step length is the collision length $\lambda_0$ rather than $X_0$ . A common design is a sandwich of iron plates and plastic scintillator. To absorb the shower you need 10–15 $\lambda_0$ , and $\lambda_0 = 17$ cm for iron.

\frac{\htmlClass{t-sig}{\sigma(E)}}{\htmlClass{t-E}{E}} = \frac{\htmlClass{t-a}{40\% - 60\%}}{\sqrt{\htmlClass{t-E}{E}\,(\text{GeV})}}

(1.125)

Same 1/√E law, three times the constant. Measuring a quark will always be worse than measuring an electron.

Every symbol, one at a time

Hover or tap a symbol above — it lights up in the equation and its meaning, units and type appear here.

🔢 Worked example — why the hadronic section dominates the building

Electromagnetic depth: 15–25 $X_0$ of lead is $15 \times 5.6 = 84$ mm to $25 \times 5.6 = 140$ mm — 8 to 14 centimetres.

Hadronic depth: 10–15 $\lambda_0$ of iron is $10 \times 17 = 170$ cm to $15 \times 17 = 255$ cm — 1.7 to 2.6 metres.

A factor of about twelve, and since the hadronic section wraps around the electromagnetic one, its volume and mass dominate the whole experiment. When you look at a photograph of a collider detector, almost all of what you see is iron whose only job is to be 17 cm per interaction.

That ratio, $\lambda_0 / X_0$ , is also why the layers go in the order they do. Lead is 5.6 mm per radiation length but about 17 cm per collision length, so a 25 $X_0$ electromagnetic calorimeter is under one interaction length thick: hadrons sail through it, leaving only a minimum-ionising trace, and get absorbed in the section built for them. The ordering is not a convention. It is forced by the two length scales of §1.11.

The complete instrument

Everything in §1.13a through this page is assembled into arrays covering the whole solid angle and pointing at the interaction region. The result is a set of concentric layers, and its power is not in any single one of them.

🧅 The layers, and what each particle does to them

light up:

Pick a particle above to light up the layers it touches, and click any layer for what it measures. No single layer identifies anything — the pattern across all five does.

A transverse slice of a generic collider experiment. Click a layer for what it measures and why it sits at that radius; pick a particle to see its signature.

⚙️ Engineer’s bridge — the detector is a one-hot decoder

Work through the widget above and a truth table falls out:

	vertex + tracker	ECAL	HCAL	muon
e±	track	shower	—	—
γ	—	shower	—	—
μ±	track	mip	mip	track
π±, K±, p	track	mip	shower	—
n, K⁰_L	—	mip	shower	—
ν	—	—	—	—

Every row is distinct. No single layer identifies anything — an ECAL shower could be an electron or a photon; an HCAL shower could be a neutron or a pion — but the combination is unique, and the decoding is a lookup, not a measurement.

Two consequences that a systems engineer will recognise immediately:

The all-zero row is a real code. A neutrino is identified by every channel reading zero, which you can only trust if the detector is hermetic: any gap in the coverage produces the same signature as a neutrino. Missing transverse momentum is the most geometry-dependent measurement in the whole of particle physics, and every discovery that rests on it (§9.6, §9.14) rests on believing the acceptance.
The tracker must not perturb what it measures. It sits before every destructive layer, so material there converts photons early and starts showers in the wrong place — corrupting the ECAL’s input. A tracker is engineered to interact as little as possible while still measuring, which is a strange and demanding specification: a sensor whose figure of merit is partly its own transparency.

Where it breaks: transparency and signal are the same knob turned opposite ways. A thinner sensor scatters less and disturbs the track less, and it also collects fewer electron–hole pairs, so the signal-to-noise falls exactly as the figure of merit you were optimising improves. There is no setting where both are free — the specification is a trade with a floor, and the floor is set by the amplifier noise the previous bridge argued you could do without.

The two length scales that lay out the whole experiment

Quantity↕	Sets↕	Typical value↕	Consequence↕
X₀ — radiation length	depth of the EM calorimeter	5.6 mm (Pb); need 15–25 X₀
λ₀ — collision length	depth of the hadronic calorimeter	17 cm (Fe); need 10–15 λ₀	1.7–2.6 m of iron — a factor twelve more, and it dominates the mass of the experiment
r_M — Molière radius	transverse cell size	16 mm (Pb), 21 mm (PbWO₄)	about 2 cm for anything dense, so every EM calorimeter ever built has cells of roughly that size
σ(E)/E	how well you measure energy	15–18 %/√E (EM), 40–60 %/√E (hadronic)	improves with energy — the only instrument in this chapter that does
βγcτ	vertex-detector resolution required	0.66 mm for a 10 GeV D⁰	10 μm silicon strips make it a 66σ measurement; 1 mm bubble-chamber optics make it impossible

Read the last column as a design brief. Nothing in the layout of a modern experiment is arbitrary — each dimension is some number of X₀, λ₀ or r_M, and those come from §1.11, which came from the physics of energy loss. <strong>The detector is the Bethe–Bloch and shower physics of this chapter, made out of metal.</strong>

Reproduce it

import numpy as np

X0Pb, X0Fe, lam0Fe = 5.6, 17.6, 170.0            # mm
print(f"Fig 1.28: 8 plates x 12.7 mm = {8*12.7:.1f} mm of Pb / X0 = {X0Pb} mm "
      f"-> {8*12.7/X0Pb:.1f} X0  (book: 18)")
print(f"EM depth  15-25 X0 of Pb   = {15*X0Pb/10:5.1f} - {25*X0Pb/10:.1f} cm")
print(f"HAD depth 10-15 lam0 of Fe = {10*lam0Fe/10:5.1f} - {15*lam0Fe/10:.1f} cm"
      f"   -> {10*lam0Fe/(25*X0Pb):.0f}x deeper")

print("Moliere radius r_M = X0 * 21.2/Ec, with Ec ~ 600/Z:")
for name, X0, Z in (("Pb", X0Pb, 82), ("Fe", X0Fe, 26)):
    Ec = 600/Z
    print(f"   {name}: Ec = {Ec:4.1f} MeV -> r_M = {X0*21.2/Ec:.1f} mm"
          f"{'     <- barely depends on the material' if name=='Fe' else ''}")

E = 100.0
print(f"resolution at E = {E:.0f} GeV: EM {15/np.sqrt(E):.1f}-{18/np.sqrt(E):.1f} %, "
      f"hadronic {40/np.sqrt(E):.1f}-{60/np.sqrt(E):.1f} %")
k, a = 0.87, 17.0                                 # tracker %/GeV, calo %*sqrt(GeV)
print(f"tracker dp/p = {k} % x p crosses the EM calorimeter at {(a/k)**(2/3):.1f} GeV")

rho, dEdx, W, d = 2.329, 1.664, 3.6, 100e-4       # g/cm3, MeV cm2/g, eV, cm
n = rho*dEdx*d*1e6/W
print(f"100 um of Si: {rho*dEdx*d*1e3:.1f} keV / {W} eV = {n:.0f} e-h pairs = "
      f"{n*1.602176634e-19*1e15:.2f} fC, no gain, sqrt(N)/N = {100/np.sqrt(n):.2f} %")
n_ar = 2.53e3/26
print(f"   argon 1 cm: {n_ar:.0f} pairs, sqrt(N)/N = {100/np.sqrt(n_ar):.1f} %, "
      f"and it still needs gain 1e5")

m, ctau, p = 1.86484, 122.9e-3, 10.0              # GeV, mm, GeV
print(f"D0 at {p:.0f} GeV: bg = {p/m:.2f}, flight = {p/m*ctau:.2f} mm = "
      f"{p/m*ctau*100:.0f} sigma at 10 um")

prints

Fig 1.28: 8 plates x 12.7 mm = 101.6 mm of Pb / X0 = 5.6 mm -> 18.1 X0  (book: 18)
EM depth  15-25 X0 of Pb   =   8.4 - 14.0 cm
HAD depth 10-15 lam0 of Fe = 170.0 - 255.0 cm   -> 12x deeper
Moliere radius r_M = X0 * 21.2/Ec, with Ec ~ 600/Z:
 Pb: Ec =  7.3 MeV -> r_M = 16.2 mm
 Fe: Ec = 23.1 MeV -> r_M = 16.2 mm     <- barely depends on the material
resolution at E = 100 GeV: EM 1.5-1.8 %, hadronic 4.0-6.0 %
tracker dp/p = 0.87 % x p crosses the EM calorimeter at 7.3 GeV
100 um of Si: 38.8 keV / 3.6 eV = 10765 e-h pairs = 1.72 fC, no gain, sqrt(N)/N = 0.96 %
 argon 1 cm: 97 pairs, sqrt(N)/N = 10.1 %, and it still needs gain 1e5
D0 at 10 GeV: bg = 5.36, flight = 0.66 mm = 66 sigma at 10 um

🔑 If you remember only three things

Vertexing turned a lifetime into a geometry. A charm particle covers a few hundred microns, so an instrument resolving tens of microns can see that it travelled — and short-lived particles stopped being inferred and started being seen.
The calorimeter is where the particle ends. Anything that needs it intact has to happen upstream, and anything downstream sees only what leaked through.
Electromagnetic showers are short and hadronic ones are long. That ratio is the whole basis of layered identification, and it is why the hadronic section is most of the building.

Where this goes next

The chapter’s problems, where all of this gets used.
§4.10 and §8.6 are the vertex detector doing its job.
§9.14 is a real experiment built out of exactly these layers, at the one luminosity where every choice on this page has to be re-argued.

✅ Check yourself — silicon, showers and the layered detector

0/6 answered · 0 correct

1.A gas chamber needs a gain of 10⁵ and a silicon strip detector needs none. Why can silicon do without?
2.A D⁰ has cτ = 123 μm and lives 0.41 ps. How does a silicon vertex detector 'measure' such a lifetime?
3.In the shower widget, take a lead calorimeter from 10 GeV to 1000 GeV. The depth of the maximum moves by only ~6.6 X₀ while σ(E)/E drops tenfold. What is the one fact behind both?
4.Why is the hadronic calorimeter placed outside the electromagnetic one, and why is it about twelve times deeper?
5.In the cutaway widget, select ν. Every layer stays dark. What makes that a usable measurement rather than no measurement?
6.The Molière radius comes out at 16 mm for lead and 16 mm for iron, and the book quotes 21 mm for lead tungstate. Why do such different materials agree so closely?