SDD vs. Si-PIN: Choosing the Right XRF Detector

Every XRF analyzer sold into a pawnshop, jewelry counter, or refinery lab contains one component that quietly decides how good the numbers on the screen will be: the detector. Two silicon technologies dominate the precious metals market — the Si-PIN diode and the Silicon Drift Detector (SDD). They look identical from the outside, they sit behind the same beryllium window, and a spec sheet will often describe both simply as “silicon detector.”

They are not equivalent. The difference between them is physical, not cosmetic, and it shows up precisely in the situations where a wrong number costs you money: a white gold alloy with rhodium plating, a 22K piece with a silver-rich solder joint, a bar with a tungsten core, a lot where palladium needs to be separated from silver.

This guide explains the physics behind that gap. It deliberately avoids brand comparisons and purchase advice — if you want help choosing a machine, our gold tester selection guide covers that ground. What follows is the engineering layer underneath: how each detector converts an X-ray photon into a number, why one resolves overlapping peaks the other cannot, and where each still makes technical sense.

What a detector actually has to do

When the analyzer’s X-ray tube irradiates a sample, atoms in that sample emit characteristic fluorescent X-rays. Gold emits at specific energies, copper at others, silver at others still. The detector’s job sounds simple: catch each incoming photon and report its energy accurately.

The mechanism is the same in both technologies. A photon enters the silicon crystal and deposits its energy by ionization, creating a cloud of electron-hole pairs. In silicon it takes roughly 3.6 eV to create one electron-hole pair, so an 8 keV copper Kα photon generates about 2,200 pairs, while a 9.7 keV gold Lα photon generates about 2,700. That charge cloud is collected, converted into a voltage pulse, and the pulse height is proportional to the original photon energy.

Everything that distinguishes a good detector from a mediocre one happens in the collection step. How completely is the charge gathered? How much electronic noise is added on the way? How quickly can the system reset and accept the next photon? Si-PIN and SDD answer those three questions differently.

The Si-PIN geometry

A Si-PIN diode is a straightforward sandwich: a p-type layer, a thick intrinsic (undoped) silicon region, and an n-type layer. Reverse bias depletes the intrinsic region and creates a uniform electric field across it. Charge generated anywhere in that volume drifts to a single large-area collecting electrode on the back face.

That large electrode is the design’s core limitation. Capacitance scales with electrode area, and the electronic noise contributed by the preamplifier scales with capacitance. A big collecting pad means a big capacitance, which means a noise floor you cannot escape by any amount of clever electronics downstream.

The SDD geometry

The Silicon Drift Detector solves the capacitance problem with a rethink of the electrode layout. Instead of one large collecting pad, an SDD uses a series of concentric ring electrodes on the surface, biased progressively to create a lateral drift field. Charge generated anywhere in the wafer is funneled sideways along that field toward a tiny central anode, often smaller than 0.1 mm².

Because the anode is minuscule, its capacitance is minuscule, and it stays essentially constant regardless of how large the overall active area is. This decoupling of active area from anode capacitance is the single most important idea in the technology. You can build a 25 mm² or 50 mm² SDD that collects far more photons than a Si-PIN of the same size while carrying a fraction of the noise penalty.

Energy resolution: the number that matters most

Energy resolution is quoted as FWHM (Full Width at Half Maximum) at the 5.9 keV manganese Kα line, measured in electron volts. It describes how narrow a peak the detector produces when fed photons of a single, precise energy. Lower is better.

Typical figures in commercial precious metals instruments:

• Si-PIN: 180–220 eV FWHM, with older or uncooled units drifting above 250 eV
• SDD: 125–140 eV FWHM, with premium units reaching 122–129 eV

A 180 eV peak is roughly 40% wider than a 130 eV peak. On a clean, single-element sample that difference is invisible — both detectors will tell you a gold coin is gold. The gap becomes decisive when two elements emit at energies close together, because two broad peaks merge into one shapeless hump while two narrow peaks stay distinguishable.

The overlaps that actually occur in precious metals

These are not hypothetical. Each of the following appears routinely on a working counter:

Tungsten Lα versus gold Lα

Tungsten Lα sits at 8.40 keV. Gold Lα sits at 9.71 keV. That is 1.31 keV of separation, which sounds comfortable, but the relevant comparison is against peak width plus the tails of the much larger gold peak. A 130 eV detector separates these cleanly and will show a distinct tungsten shoulder when tungsten is present near the surface. A 200 eV detector smears the region, and a small tungsten contribution can hide inside the skirt of the gold peak.

This matters for counterfeit screening, though it is worth stating the limitation plainly: no detector, however good, sees a tungsten core buried under 1 mm of gold, because XRF only interrogates the top few tens of microns. Resolution helps you catch surface and near-surface tungsten, not deep cores. Our tungsten detection tutorial covers the layered verification methods required for the deep case.

Rhodium Lα versus palladium Lα versus silver Lα

Rhodium Lα is at 2.70 keV, palladium Lα at 2.84 keV, silver Lα at 2.98 keV. These sit within 140 eV of each other — meaning the total spread across all three is about the same as the FWHM of a single good SDD peak, and considerably narrower than a single Si-PIN peak.

A Si-PIN cannot resolve this cluster in the L series at all; it sees one broad bump. This is a live problem for white gold, which is frequently rhodium plated over a palladium-bearing alloy, and for anyone handling PGM-containing material.

Lead Lα versus arsenic Kα

Lead Lα at 10.55 keV and arsenic Kα at 10.54 keV are functionally coincident. No detector separates them by energy alone; both require deconvolution using secondary lines. But the quality of that deconvolution depends entirely on how cleanly the software can measure the secondary peaks, which brings you back to resolution.

Count rate and dead time

Resolution is the headline number, but throughput determines how long an operator waits.

Every time a photon is processed, the electronics are occupied for a short interval and cannot accept another. That interval is dead time. If photons arrive faster than the system can process them, two arriving close together get summed into a single false pulse at the combined energy — a pile-up peak, which is a genuine artifact that can be misread as a real element.

Because an SDD’s small anode allows much shorter shaping times without a noise penalty, SDDs handle dramatically higher input rates:

• Si-PIN: useful throughput typically caps around 10,000–20,000 counts per second
• SDD: comfortably handles 100,000–1,000,000 counts per second depending on the model

The practical consequence is measurement time. To reach a given statistical precision you need a certain total number of counts. Collecting 500,000 counts at 15,000 cps takes over 30 seconds; at 200,000 cps it takes under 3 seconds. On a counter processing a queue of customers, or a refinery grading an incoming lot piece by piece, that ratio compounds through the whole day.

There is a second, subtler benefit. Shorter measurement times mean less thermal drift within a single reading and less opportunity for a handheld or loosely positioned sample to shift mid-measurement.

Cooling, stability, and drift

Silicon detectors generate thermally excited charge carriers — leakage current — that add noise indistinguishable from signal. Leakage current roughly doubles for every 7°C rise, so temperature control is not optional.

Both detector types use thermoelectric (Peltier) cooling rather than the liquid nitrogen required by older lithium-drifted germanium systems. The difference lies in how much cooling is required to reach the target performance and how stable that state is.

Si-PIN devices, carrying more intrinsic noise, depend more heavily on cooling to reach their rated specification. They generally need longer to reach thermal equilibrium and show more resolution degradation when ambient conditions swing. A shop that runs hot in the afternoon, or a machine positioned near a window or HVAC outlet, will see measurable drift.

SDDs reach their rated resolution with less aggressive cooling and hold it more consistently across an ambient range. This is a durability argument as much as a performance one: a detector that spends less time at extreme thermal gradients tends to hold calibration longer between service intervals.

In either case, the single most common cause of “the machine suddenly reads wrong” is an instrument used before it has finished warming up. Thirty minutes from cold start is a reasonable rule regardless of detector type. Our calibration guide covers the verification routine that should follow warm-up.

What the spectrum looks like in practice

Specifications describe capability in the abstract. What an operator actually sees is a spectrum, and learning to read it is more useful than memorizing FWHM figures.

A clean 18K yellow gold reading

On plain 18K yellow gold — nominally 75% Au, with copper and silver making up the balance — both detector types produce a recognizable result. Gold Lα dominates at 9.71 keV, gold Lβ appears at 11.44 keV, copper Kα sits at 8.05 keV, and silver Kα at 22.16 keV.

These are widely separated. A Si-PIN handles this case without difficulty, and anyone whose entire business is plain yellow karat gold should understand that the resolution argument does not apply strongly to them. The peaks are far enough apart that 200 eV is ample.

The same alloy with rhodium plating

Now add a rhodium flash, standard on white gold and increasingly common on yellow pieces sold as tarnish resistant. Rhodium Lα appears at 2.70 keV, in the crowded low-energy region alongside palladium and silver L lines.

On an SDD the rhodium contribution is a discrete peak, and because rhodium has no business being in the bulk alloy, its presence immediately signals a surface treatment. On a Si-PIN the low-energy region is a single unresolved rise, and the rhodium signature is not separable. The instrument may still return a plausible gold percentage — which is precisely the danger, because a plausible wrong number is worse than an obvious failure.

Reading the shape, not just the number

The habit worth building is checking peak shape before accepting a composition result. A genuine single-element peak is close to symmetric. A peak with a pronounced shoulder, an unexplained asymmetry, or a flat top is usually two unresolved peaks in disguise.

Operators who only ever look at the composition table never see this. Operators who glance at the spectrum catch problems the software’s automatic quantification silently smoothed over. This is a training issue rather than a hardware issue, and it applies to both detector types — though on a Si-PIN it happens far more often.

Cost, service life, and total ownership

Detector choice affects the ownership cost of an instrument in ways that are not obvious at purchase.

The replacement question

Detectors are consumable on a long timescale. Window contamination, thermal cycling fatigue, and gradual charge-collection degradation all accumulate. An SDD assembly costs meaningfully more to replace than a Si-PIN assembly, and this is a legitimate point in Si-PIN’s favor for a low-stakes application.

However, the comparison should be made against the cost of the errors avoided, not against the hardware price alone. A single misidentified plated item or one mis-settled lot can exceed the lifetime price difference between the two technologies. The seven-year ownership analysis works through that arithmetic across a realistic transaction volume.

What actually shortens detector life

In field experience the dominant causes of premature detector failure have nothing to do with which technology is inside:

• Physical damage to the beryllium window from a sample dropped or dragged across it. The window is thin and unforgiving.
• Repeated power cycling that subjects the cooler to constant thermal swings. Leaving an instrument powered during business hours is gentler than switching it off between customers.
• Dust and metal fines accumulating in the sample chamber, particularly in scrap and refinery environments where filings are constantly present.
• Ambient temperature extremes, especially instruments positioned in direct sunlight or immediately beside an HVAC outlet.

Addressing these four factors extends service life more than any detector selection decision. They are covered as part of the deployment checklist in our maintenance and calibration reference.

Detection limits and light elements

Limit of detection (LOD) is where resolution and count rate compound into a single practical outcome. LOD improves with the square root of counting time and degrades with background noise. An SDD’s narrower peaks concentrate signal into fewer channels while the background stays flat, raising the peak-to-background ratio, and its higher count rate accumulates statistics faster.

For mid-Z elements in a gold matrix, the practical difference in a working instrument:

• Si-PIN: reliable quantification generally down to roughly 0.1–0.5% for most alloying elements
• SDD: reliable quantification down to roughly 0.01–0.05%, i.e. 100–500 ppm

That order of magnitude decides whether you can see trace iridium in a bar, quantify low-level cadmium in a solder alloy, or detect the small nickel fraction that determines whether a piece is EU nickel-release compliant.

Where both technologies genuinely struggle

It is worth being honest about a shared limitation rather than presenting SDD as universally capable. Elements lighter than roughly sodium (Z=11) emit X-rays so low in energy that they are absorbed by air and by the detector window before they can be counted. Even magnesium, aluminium, and silicon are difficult without a vacuum or helium purge that benchtop precious metals instruments rarely include.

In practical terms: neither detector will tell you about the organic residue, the lacquer coating, or the epoxy fill in a hollow piece. Both will report the metal fraction and silently ignore everything below the light-element cutoff. An operator who forgets this will over-report gold content on a filled bangle — a failure mode covered in our 22K wedding gold tutorial.

Reading a spec sheet without being misled

Detector specifications are quoted under favorable conditions that rarely match a working counter. A few things to check before accepting a number:

Ask at what count rate the resolution was measured

Resolution degrades as count rate rises. A detector quoted at 129 eV may deliver 129 eV only at low input rates and 145 eV at the rates you will actually use. A trustworthy specification states both figures, or states the rate at which the resolution number applies.

Ask about active area alongside resolution

A very small SDD can post an excellent FWHM while collecting so few photons that measurement times become long. Resolution and active area (typically quoted in mm²) must be read together — 25 mm² at 135 eV can outperform 7 mm² at 125 eV for real throughput.

Treat “silicon detector” as an unanswered question

The phrase covers both technologies. If a datasheet says “high-performance silicon detector” without naming SDD or Si-PIN and without an FWHM figure in eV, assume Si-PIN until proven otherwise, and ask directly.

Check whether the FWHM is measured or theoretical

Some specifications quote the detector chip’s intrinsic resolution rather than the assembled system’s. The system figure — detector plus preamplifier plus pulse processor plus cooling as installed — is the only one that predicts your results.

When Si-PIN is still the right technical choice

Presenting SDD as the answer to every question would be dishonest. Si-PIN remains defensible in specific circumstances:

• Single-alloy, high-volume sorting where the only question is a pass/fail on one well-separated element, and no overlapping peaks are in play.
• Legacy fleet consistency, where a business already runs several Si-PIN units with an established empirical calibration and mixing detector types would introduce systematic offsets between stations.
• Severe budget constraint where the alternative is not an SDD but no instrument at all — a Si-PIN reading is enormously better than a touchstone guess.
• Simple karat confirmation on plain yellow gold with no plating question, where 0.2–0.5% precision is entirely adequate.

The honest summary is that Si-PIN answers “is this roughly what it claims to be?” reliably, while SDD answers “exactly what is in this, including the parts someone hoped I would miss?”

Matching detector capability to your actual risk

The useful question is not which detector is better in the abstract, but which failure mode would hurt you most.

If your exposure is counterfeit and fraud

You need resolution. Tungsten shoulders, plating signatures, and unexpected trace elements are all resolution-limited detections. A 200 eV detector will pass material that a 130 eV detector flags.

If your exposure is settlement accuracy

You need both resolution and low LOD. When you are paying by assay, a systematic 0.3% error on every transaction is a permanent margin leak that compounds silently across a year of volume. This is the arithmetic behind our ROI model.

If your exposure is throughput

You need count rate. A queue of customers at a busy counter, or a large incoming lot graded piece by piece, turns a 30-second measurement into an operational bottleneck.

If your exposure is regulatory

You need low LOD and documented traceability. Nickel release, cadmium restriction, and hallmarking compliance all depend on quantifying elements at levels a Si-PIN may not resolve from background.

A short field diagnostic

If you already own an instrument and are unsure which detector is inside it, three checks will usually tell you without opening anything:

1. Measure a known white gold piece with rhodium plating. If the spectrum shows a single broad bump in the 2.7–3.0 keV region rather than distinguishable Rh, Pd, and Ag features, you are looking at Si-PIN behavior.
2. Time a standard measurement to a fixed count target. If reaching acceptable statistics on a small item consistently takes more than 15–20 seconds, throughput is Si-PIN-class.
3. Compare a cold-start reading to one taken 40 minutes later on the same reference standard. A shift greater than a few tenths of a percent points to a detector more dependent on thermal equilibrium.

None of these are definitive on their own, but together they are a reliable indication. If your results suggest limitations you are not comfortable with, the setup and deployment checklist covers environmental fixes that recover some performance before replacement becomes necessary.

Frequently asked questions

Is a lower FWHM number always better?

Lower FWHM means better energy resolution, so in isolation yes. But it must be read together with active area and the count rate at which it was measured. A very small detector can post an excellent FWHM while collecting too few photons for fast, statistically sound measurements.

Can an SDD detect a tungsten core in a gold bar?

Not if the core is buried. XRF interrogates only the top few tens of microns of a sample regardless of detector quality. An SDD is substantially better at spotting tungsten at or near the surface because it resolves the W Lα peak at 8.40 keV from the gold Lα at 9.71 keV, but deep cores require density measurement, ultrasonic testing, or destructive sampling.

Does an SDD need liquid nitrogen cooling?

No. Both SDD and Si-PIN detectors in modern precious metals instruments use thermoelectric Peltier cooling. Liquid nitrogen was required by older lithium-drifted germanium and silicon detectors, which are not used in this class of instrument.

Why does my analyzer report different results for the same item?

The most common causes are insufficient warm-up, inconsistent sample positioning, and testing a plated or non-homogeneous item at different spots. Detector type affects the magnitude of scatter but not these root causes. Allow 30 minutes from cold start and use a consistent multi-spot protocol before concluding the instrument is at fault.

What does “dead time” mean on my display?

Dead time is the percentage of measurement time during which the electronics were busy processing a photon and unable to accept another. Moderate dead time is normal. Very high dead time indicates the input rate is saturating the processor, which can produce pile-up artifacts — false peaks at the summed energy of two coincident photons.

Can I upgrade the detector in an existing analyzer?

Generally not as a field operation. The detector, preamplifier, pulse processing electronics, cooling assembly, and calibration model are engineered as a matched system. Replacing the detector alone requires re-engineering the signal chain and a full recalibration, which in most cases costs more than it saves.

How much does resolution degrade as a detector ages?

A well-maintained detector in a stable environment degrades slowly, often remaining within specification for many years. Degradation accelerates with thermal cycling, mechanical shock, and window contamination. Tracking FWHM on a reference standard at fixed intervals is the most reliable early warning of a developing problem.

Does detector type change how often I need to calibrate?

Calibration frequency is driven mainly by usage volume, environmental stability, and the accuracy tolerance your application requires. Detector type influences drift behavior — Si-PIN units tend to be more thermally sensitive — but does not replace a scheduled verification routine using certified reference materials.

The short version

The physical difference is anode size. An SDD funnels charge to a tiny central anode, decoupling capacitance from active area; a Si-PIN collects on a single large electrode and pays a permanent noise penalty for it.

That one design choice produces everything else: roughly 130 eV versus 200 eV resolution, an order of magnitude more count rate, an order of magnitude lower detection limits, and better thermal stability.

Whether that gap matters depends entirely on what you test. Plain yellow gold of known provenance does not require it. Plated white gold, PGM-bearing material, suspected counterfeits, settlement-grade assay, and regulatory compliance all do. The right question is not “which detector is better” but “what am I currently unable to see, and what would it cost me to keep not seeing it?”

For the underlying physics of how these fluorescent X-rays are produced in the first place, see our guide to the physics of XRF alloy composition. For the measurement-depth constraints that limit every XRF instrument regardless of detector, see plating thickness measurement limits.