Avalanche Photodiode

If a small load resistance R_Lis used to increase the frequency bandwidth of a PIN photo-diode, the signal voltage may be quite small, requiring amplification. This can be accomplished with electronic amplifiers, but these introduce their own sources of noise, and it is sometimes desirable to increase the signal generated by the detector, before amplification. One way to do this is through the avalanche multiplication process, depicted in Fig. 14-13a. This is the solid-state analog of the electron multiplication that takes place in a pho-tomultiplier tube. An electron is accelerated by the E field in the depletion region of a reverse-biased p-n junction, and gains kinetic energy in proportion to the distance traveled. When the electron's kinetic energy is high enough, it can collide with an atom in the semiconductor and create an additional electron-hole pair, a process termed impact ion-

Figure 14-13 (a) An electron accelerated in a strong E field creates an additional electron-hole pair by impact ionization. Both the new electron and the original one then create additional electron-hole pairs, resulting in avalanche multiplication. (b) Impact ionization can occur when an electron in the CB picks up kinetic energy greater than the band-gap energy.

ization. There are now two electrons, and as each one of these accelerates in the E field, it can create yet another electron-hole pair by the same mechanism. There are now a total of four electrons, each of which can create another one to give eight, and so on. The result is avalanche multiplication, in which the number of charge carriers increases exponentially with distance traveled.

The creation of electron-hole pairs by impact ionization can be understood in terms of the energy band picture of Fig. 14-13b. After moving a distance x in a direction opposite to the E field, the electron loses an amount of potential energy of magnitude eE x. If no energy is lost to other processes, the electron then gains this same amount of kinetic energy K. When K > E_g, it becomes energetically possible for a collisional-energy transfer process to take place, in which the electron kinetic energy decreases by K = -E_g, while at the same time the potential energy of a valence electron is increased by this same amount. Increasing the potential energy of a valence electron by an amount E_gcorresponds, in the band picture, to taking an electron out of the valence band and placing it in the conduction band, that is, creation of an electron-hole pair.

The energy needed to create an electron-hole pair is usually much less than the potential energy change of an electron as it moves across a reverse-biased p-n junction. For example, a typical band-gap energy is E_g ~ 1-2 eV, whereas a 10 V reverse bias corresponds to a potential energy change of 10 eV. Since the electron acquires sufficient energy to create several electron-hole pairs as it moves through this potential difference, it might seem that avalanche multiplication should occur readily in most reverse-biased photodiodes. However, the avalanche phenomenon actually plays a minor role in conventional PIN photodiodes. The explanation for this is that the electron undergoes nonionizing collisions as well as ionizing collisions as it moves through the electric field. For example, the electron can scatter off the thermally induced vibrations in the material (lattice phonons), giving up kinetic energy to heat. At moderate electric field values, these nonionizing collisions prevent the electron's kinetic energy from reaching the threshold value K = E_g. When the electric field is sufficiently high, however, the electron can pick up kinetic energy K > E_gbefore a nonionizing collision occurs, and the avalanche mechanism becomes more efficient.

The electric field at which the avalanche mechanism becomes important depends on the material, being higher in wider-band-gap materials that have a higher threshold kinetic energy. For silicon, an electric field E ~ 5 × 10⁷ V/m over a path length of ~ 2 m is needed for efficient avalanche multiplication. This corresponds to a potential difference V~ 100 V, much higher than the reverse bias of a typical PIN photodiode.

A photodiode utilizing avalanche multiplication to achieve gain is termed an avalanche photodiode, or APD. The structure of an APD, depicted in Fig. 14-14, differs from that of the PIN photodiode (Fig. 14-12) in two ways. First, light enters through a

Figure 14-14 In an avalanche photodiode (APD), electrons photoexcited in a nearly intrinsic region are swept out by a small electric field there, and injected into a high-field region between highly doped n and p layers. Avalanche multiplication occurs primarily in the high-field region.

highly doped n-type layer rather than a p-type layer. Second, an additional p-type layer has been added between the highly doped n layer and the nearly intrinsic layer. The leftmost n and p layers are highly (and nearly equally) doped, so the junction width between them is small (d ~ 2 m), and the high electric field is mostly confined to this region. In the adjacent intrinsic region (actually lightly doped p-type), there is a much smaller and nearly uniform E field, which extends out to the highly doped p⁺ region on the right.

The APD operates in the following way. Light passes through the thin n⁺ and p layers and is absorbed in the much thicker intrinsic region. Electrons and holes created by pho-toabsorption then drift in opposite directions under the influence of the E field, electrons to the left and holes to the right. The electrons eventually make it to the high-field region, where they undergo avalanche multiplication. Holes do not initiate the avalanche in this scheme, but those created by impact ionization can contribute to its development. In silicon, however, holes are much less efficient at causing ionization than are electrons, and therefore make only a minor contribution to the amplification.

In principle, an APD could be constructed with a p⁺-n-i-n⁺ structure, instead of the n⁺-p-i-p⁺ shown in Fig. 14-14. However, in this case it would be the holes that are injected into the high-field region, and in silicon this would result in weak amplification. For this reason the n⁺-p-i-p⁺ structure is always used for silicon APD's. This asymmetry between electron and hole ionization probabilities also has implications for the signal-to-noise properties of the APD. Since avalanche multiplication is a statistical process, there is less statistical variation in output current (i.e., less noise) when only one type of charge carrier contributes to the avalanche. For this reason, germanium APDs, in which the electrons and holes have nearly equal ionization probabilities, are inherently more noisy than silicon APDs.

The effect of avalanche multiplication can be characterized by the multiplication factor M, defined as the ratio of photocurrent with amplification to photocurrent without amplification. The detector output voltage is then still given by Eq. (14-19), with Eq. (14-18) replaced by

M can be as high as 100 in a silicon APD, but is more typically ~ 10 in a germanium APD.

The operation of the APD depends on the proper electric field profile within the device, and this in turn requires the proper bias voltage. The relation between E(x) and bias voltage (magnitudes only) is

where we have neglected the built-in potential V₀compared with V_B. As the bias voltage increases, the area under the E(x) curve increases proportionately, as illustrated in Fig. 14-15a. At some critical voltage, the depletion region "reaches through" to the highly doped p⁺ region, with the electric field extending uniformly throughout the intrinsic region. Electrons generated anywhere in the intrinsic region are then efficiently swept out and injected into the high-field region for amplification. A device biased in this way is termed a reach-through APD.

Figure 14-15 (a) The area under E(x) increases with V_Buntil the field "reaches through" the intrinsic region. (b) Photocurrent i versus reverse-bias voltage for three values of incident light power P_in, assuming = 10 mA/mW. In the linear regime (after reach-through), i = P_in, but in the Geiger regime i becomes independent of P_in.

Fig. 14-15b shows a typical variation of responsivity with applied bias voltage for a reach-through APD. Below the threshold value (usually ~ 100-200 V), increases with V_Bas the E field starts to extend into the intrinsic region. There is a linear operating region above this, in which all photogenerated electrons are collected for amplification. In this region, the avalanche is well behaved, and the detector output is proportional to the incident light power. At still higher bias voltage, the avalanche process becomes uncontrolled, and avalanche breakdown ensues. In this situation, the detector output rises to a saturation value which is independent of the number of photons absorbed. This is the Geiger mode regime, analogous in operation to the Geiger counter used to detect nuclear radiation. The APD Geiger mode is useful when the purpose is to determine whether any photons are present, rather than to determine their number. It has applications in photon counting, in which the arrival time of individual photons is measured.

The responsivity of an APD generally decreases with increasing temperature, due to the increasing probability of electrons making nonionizing collisions with phonons. These collisions take away some of the kinetic energy gained by an electron that might otherwise be available for creating additional electron-hole pairs. In order to stabilize the gain of APD detectors in a changing ambient temperature, then, temperature-control circuitry is needed. Despite these complications, and the need for high-voltage bias, the improved responsivity of the APD makes it an attractive choice for applications limited by a weak light signal, such as in fiber optic communications.