Silicon-based photon counters can accommodate both the detector and the counting electronics into a single, compact package. Image courtesy of sensL.
Stephen Mounsey investigates photon counting, and discovers a rich variety of approaches to an old problem
When a piece of metal absorbs light of a frequency above a certain threshold value, (corresponding to UV wavelengths for most metals), it gains a charge, as some of its electrons are removed. This is called the photoelectric effect, and after it was first described mathematically by Einstein in 1905, it went on to form the basis of our understanding of the way in which light contains quanta of energy – what we now call photons.
This photoelectric effect has been exploited for many years in the workings of photomultiplier tubes (PMTs), which make use of the effect in order to detect light at very low intensities, even corresponding to single photons. These analogue devices incorporate a photoelectrode within a vacuum tube, with a ladder of several secondary electrodes – called dynodes – leading to an output anode. Each of the dynodes is held at a high voltage compared to the photoelectrode, and so any electrons displaced by the incident photon are accelerated by a strong electric field, causing a cascade of ever-increasing intensity from dynode to dynode, and a net gain to the detectable signal at the anode; the higher the current at the anode, the more photons are being detected by the device.
Photon counting refers to applications wherein knowledge of individual photon detection events is required; rather than simply detecting a low-intensity signal, a photon counting application may need to know, as the name suggests, precisely how many counting events have taken place within a given time duration. Uwe Ortmann is head of sales and marketing at PicoQuant, a German company specialising in electronics for photon counting. He describes the way photon counting differs from analogue low-light detection: ‘When using analogue signals, you may need tens, or even hundreds of photons in order to get a measurable signal, but when using true photon counting, a single photon should be detectable.’
Typically, a modern PMT is capable of emitting a signal of one million electrons when it absorbs a single photon. If, however, the gain was 100 times lower, the signal would not be noticeable over the background noise of the system. Tim Stokes, general sales manager at Hamamatsu, explains that for a PMT to be useful for photon counting, it therefore needs to fulfil two criteria: high gain, and low noise (sometimes called dark count). In practice, the most difficult applications at which to achieve this low dark count are those in the infrared region of the spectrum, as noise can result from heat leaking in from any warm matter around the detector. Cooling of detectors leads to lower noise and higher sensitivity, and Hamamatsu produces PMTs capable of single photon performance up to 1.7micron, albeit with cryogenic cooling.
Every photon detector can be given a property known as the quantum efficiency, or QE. This number describes the efficiency with which a given photoelectrode transmutes the energy of a photon into energetic electrons. A high QE photoelectrode means a high likelihood of a given incident photon producing a detectable signal, which in turn influences the efficiency of the detector as a whole. Some advanced PMTs make use of InGaS or InGaAs as the photocathode, and these materials allow the detector to achieve a higher QE. Higher QE does, however, mean more noise, and so these effects are traded off against each other to get the best possible results on an application-specific basis. Stokes says that some visible light PMTs can have QEs of up to 0.5 when using photoelectrodes made of GaAsP. As Ortmann from PicoQuant says: ‘Not every photon will be converted to an electronic signal, but what is important is that every converted photon will be detected by the counting electronics.’
PMTs are produced on a small scale, often by hand, and they are expensive. Although the technology is old, PMTs are still widely used because they are able to cover a wide area per device, and they can have the lowest noise of any photon detecting technology. The gain of an analogue PMT is linearly proportional to the voltage difference between the photocathode and the signal anode.
When more detectors are needed in bulk applications or an all-digital solution is desirable, or when conditions prohibit the use of delicate PMTs, digital devices based on photodiodes are used. Similarly to a PMT, a photodiode contains a junction with a voltage difference across it. Incident photons displace charge (in this case, the charge carriers are lattice defects in the semiconductor) allowing a current to flow across the junction, and a signal to be detected. As is the case with PMTs, the gain in the signal is proportional to the voltage, but in the case of a photodiode, the operating voltage is limited by the breakdown potential of the semiconductor. Silicon is the main material used in making semiconductor detectors, and the simplest type of low-intensity detector is called the avalanche photodiode. The gain obtainable by a simple APD is in the order of 1,000x, but as mentioned, in order to get a signal above background for single photon counting, you need a gain of around a million.
Tim Stokes from Hamamatsu explains that ‘tricks’ are used in order to get higher gain. ‘The semiconductor junction within the APD actually breaks down at a certain voltage, leading to a huge spike of noise. The technique used sees that voltage maintained at just above that breakdown potential, and as soon as the detector absorbs a photon, that huge spike is readily detectable.’ Once the semiconductor has absorbed a photon, and broken down in this way, the voltage is reduced back to a level below the breakdown threshold in order to ‘quench the APD’, before being raised back over threshold, ready to see the next event. Single signal spikes in this regime therefore correspond to single photon absorptions.
In practice, the quenching requires complicated control electronics, and it takes a certain amount of time during which the detector will be effectively blind. An advantage of the approach is that silicon has high QE, meaning that fewer photons will be missed and that signal-to-noise ratio is high. Danielle Moraes from iD-Quantique, a company specialising in SPADs (single photon avalanche diodes), states that QE refers to the material of the photodiode itself, but that the detector efficiency is what’s important for the detector as a whole. ‘SPADs either count or they do not count,’ says Moraes, ‘and therefore the device efficiency is lower than it would be in analogue devices such as PMTs.’
Silicon-based detectors tend to have a higher noise level than PMTs, and the detectors are made as small as possible in order to minimise this noise, usually somewhere in the region of 100MIC or less: ‘That’s all fine if you can focus all of the incident light into such a tiny detector,’ says Stokes, ‘but to make these devices more practical for use in commercial applications, without having to design horrendously expensive optics, the technology has leant itself to developing multi-pixel detectors.’ A typical multi-pixel detector has several thousand of these 100MIC pixels on a single chip, making the chip several millimetres in scale. At this size, the detector is reasonably able to detect light without complicated optics. Although a single detector contains many photodiodes, it is still designed to give a single output signal. Each of these 3,000-4,000 SPADs acts independently, and each has its own on-chip quenching circuitry.
A 4-side scalable silicon photomultiplier, from sensL.
Because of the increased size, and single-photon sensitivity, these multi-pixel devices are sometimes called silicon photomultipliers, or SiPMs. The downside to this approach is that dark counts are still higher than those obtained with a cooled PMT, and the devices are still only a few millimetres across, compared to PMTs, which can be 20-30cm across with a noise performance that is less dependent on size. The QE of the devices remains their chief advantage, along with the robustness and mass-predictability inherent to solid-state technology. SPADs operate at low voltages (around 5V compared to several hundred volts for PMTs). Irish company sensL is developing this technology further by extending the technique to larger arrays of SiPMs, with a broader range of applications. Carl Jackson, chief technology officer at sensL, says that by combining many photodiodes, it becomes possible to reduce the down time between detections. Instead of the timing electronics being separate to the array, photon counting arrays may have integrated read-out circuits, allowing a larger array to retain the fine-scale time resolution one might expect from a single photodiode. sensL claims a timing precision of better than 200ps for some of its arrays.
Perkin Elmer has also been investing in SiPM technology, having acquired a licence to develop it from the Max Planck institute in Munich, Germany. The company is looking to use well-established semiconductor device fabrication techniques in order to produce high-quality devices in cost-effective volumes. Christoph Witte, product manager for photon counting at the company, says that ‘the future is not a matter of SPADs vs SiPMs; the SiPM is a logical development of the APD.’
SiPMs aim to replace conventional PMTs in applications such as fluorescence detection and positron emission tomography (PET scanning). PET scanning is used to build up an accurate image of a patient’s brain, complete with information about which parts of the brain are consuming glucose (and are therefore active) on a real-time basis. The technique involves injecting the patient with an isotope-enriched form of glucose, which releases an unstable oxygen isotope when it is consumed by the brain. This oxygen isotope decays quickly within the brain to form a positron – the antimatter counterpart to an electron – which immediately annihilates with an electron from the brain, releasing energy in the form of high energy photons of gamma radiation, which exit in exactly opposite directions, allowing their origins to be traced precisely by way of detection and timing apparatus accurate enough to resolve them on an individual basis. Currently, PET scanners rely on PMTs to resolve pairs of gamma-ray photons, as they provide a wide coverage per device and because each PET scanner needs around 0.5m² of detection area.
SiPMs can be used in PET scanners by adding a layer of a dense scintillating material between the patient and the detectors. This material absorbs the high-energy gamma photon, and re-releases it as one or several photons of visible light, which in turn are detected by the SiPMs. In order to be competitive, sensL’s Carl Jackson believes that a silicon-based detector would require around 200m individual photodiodes. Jackson states that the most up-to-date products the company is shipping have 5.8m detectors on an array for smaller organ specific PET scanners, and so it has some way to go in working towards its goal of an all-digital, commercial grade full body PET scanner. However, aside from the cost advantages of a solid-state solution, doing away with the PMTs may also have a diagnostic benefit; because PMTs incorporate streams of electrons, the devices are sensitive to magnetic fields and cannot be used near MRI scanners. sensL has developed detector arrays that are robust in magnetic fields up to a high 10-Tesla – higher than fields used in most MRI scanners. This could allow PET and MRI images to be taken at the same time. Jackson states that the technique is in the process of being tested by Samsung Medical in Korea, and by a group at Stanford University in the US.
For some applications it is not sufficient to merely gain a knowledge of how many photons have been detected, but it is also important to know the precise time index at which they were detected. Fluorescence lifetime analysis can be used to determine the chemical environment around a given fluorophore, usually in biochemical circumstances. A test molecule such as a protein may be marked with a compound that fluoresces under a certain wavelength of light. These fluorophores are excited by laser radiation of a specific wavelength, and then a single-photon detector of some form (usually a SPAD) detects the photon emitted when the excitement falls back down to the ground state. By having a system capable of resolving the precise time at which the signal is received, detailed information can be built up about the lifetime of the fluorescence under various circumstances.
This application depends not only on a short time passing between the absorption of the photon and the emission of the signal, but also on the detector having a low ‘jitter’, or uncertainty in the signal. In practice, the fastest detectors are the micro-channel photomultipliers, or MCPs. These are similar in operation to the analogue PMTs, in that they rely upon a high voltage to produce gain, and they produce an analogue signal. MCPs, however, are monolithic tubes in doped glass, produced on a very small scale, which allows the detectors to respond more quickly than any other detector technology.
Witte, of Perkin Elmer, which specialises in channel photomultipliers, states that an MCP can count single photons at speeds of up to 25m per second, but adds that this isn’t a very high frequency compared to other technologies. A new development, the gigahertz photon detection module (GDPM), is suitable for higher signal levels, while retaining the low noise characteristics and rapid response required for time-resolved measurements.
In the end, picking the right photon detector for the job depends on the requirements of the application. A detector that does well in one sense is likely to have drawbacks in another, so it is important that the designer knows exactly which property is most important for the success of the application.