High-Frequency Amplifiers:Optical Low-Noise Preamplifiers

Optical Low-Noise Preamplifiers

Figure 24.22 shows a simple schematic diagram of an optical receiver, consisting of a photodetector, a preampliﬁer, a wide-band voltage ampliﬁer, and a predetection ﬁlter. Since the front-end transimpedance preampliﬁer is critical in determining the overall receiver performance, it should possess a wide band- width so as not to distort the received signal, high gain to reject noise from subsequent stages, low noise to achieve high sensitivity, wide dynamic range, and low intersymbol interference (ISI).

Front-End Noise Sources

Receiver noise is dominated by two main noise sources: the detector (PIN photodiode) noise and the ampliﬁer noise. Figure 24.22 illustrates the noise-equivalent circuit of the optical receiver.

PIN Photodiode Noise

The noise generated by a PIN photodiode arises mainly from three shot noises contributions: quantum noise Sq(f), thermally generated dark-current shot noise SD(f) and surface leakage-current shot noise SL(f). Other noise sources in a PIN photodiode such as series resistor noise are negligible in comparison. The quantum noise Sq(f), also called signal-dependent shot noise, is produced by the light-generating nature of photonic detection and has a spectral density Sq(f) = 2qIpd Df, where Ipd is the mean signal current arising from the Poisson statistics. The dark-current shot noise SD(f) arises in the photodiode bulk material. Even when there is no incident optical power, a small reverse leakage current still ﬂows resulting in shot noise with a spectral density SD(f) = 2qIDB Df, where IDB is the mean thermally generated dark current. The leakage shot noise SL(f) occurs because of surface effects around the active region, and is described by SL(f) = 2qISL Df, where ISL is the mean surface leakage current.

Amplifier Noise

For a simple noise analysis, the pre- and postampliﬁers in Figure 24.22 are merged to a single ampliﬁer with a transfer function of Av(w). The input impedance of the ampliﬁer is modeled as a parallel combi- nation of Rin and Cin.

If the photodiode noise is negligibly small, the ampliﬁer noise will dominate the whole receiver noise performance as can be inferred from Figure 24.23. The equivalent noise current and voltage spectral densities of the ampliﬁer are represented as Si(A2/Hz) and Sv(V 2/Hz), respectively.

Resistor Noise

The thermal noise generated by a resistor is directly proportional to the absolute temperature T and is represented by a series noise voltage generator or by a shunt noise current generator [21] of value

where Cpd is the photodiode capacitance, and Rin and Cin the input resistance and capacitance of the ampliﬁer, respectively. Assuming that the photodiode noise contributions are negligible and that the ampliﬁer noise sources are uncorrelated, the equivalent input noise current spectral density can be derived from Figure 24.23 as

where B is the operating bit-rate and I2 (= 0.56) and I3 (= 0.083) the Personick second and third integrals, respectively, as given in Ref. [22].

According to Morikoni et al. [23], the Personick integral in Eq. (24.35) is correct only if a receiver produces a raised-cosine output response from a rectangular input signal at the cutoff bit rate above which the frequency response of the receiver is zero. However, the Personick integration method is generally preferred when comparing the noise (or sensitivity) performance of different ampliﬁers.

Optical Sensitivity

Optical sensitivity is deﬁned as the minimum received optical power incident on a perfectly efﬁcient photodiode connected to the ampliﬁer, such that the presence of the ampliﬁer noise corrupts on average only one bit per 109 bits of incoming data. Therefore, a detected power greater than the sensitivity level guarantees system operation at the desired performance. The optical sensitivity is predicted theoretically by calculating the equivalent input noise spectral density of the receiver [24], and is calculated as

Since the number of photogenerated electrons in a single bit is very large (more than 104) for optoelectronic- integrated receivers [25], Gaussian statistics of the above BER equation can be used to describe the detection probability in PIN photodiodes.

SNR at the Photodiode Terminal

Among the photodiode noise sources, quantum noise is generally dominant and can be estimated as

When a pulse passes through a band-limited channel, it gradually disperses. When the channel band- width is close to the signal bandwidth, the expanded rise and fall times of the pulse signal will cause successive pulses to overlap, deteriorating the system performance and giving higher error rates. This pulse overlapping is known as ISI. Even with raised-signal power levels, the error performance cannot be improved [26].

In digital optical communication systems, sampling at the output must occur at the point of maximum signal to achieve the minimum error rate. The output pulse shape should therefore be chosen to maximize the pulse amplitude at the sampling instant and should give a zero at other sampling points, i.e., at multiples of 1/B, where B is the data rate. Although the best choice for this purpose is the sinc-function pulse, in practice, a raised-cosine spectrum pulse is used instead. This is because the sinc-function pulse is very sensitive to changes in the input pulse shape and variations in component values, and because it is impossible to generate an ideal sinc function.

Dynamic Range

The dynamic range of an optical receiver quantiﬁes the range of detected power levels within which correct system operation is guaranteed. Dynamic range is conventionally deﬁned as the difference between the minimum input power (which determines sensitivity) and the maximum input power (limited by overload level). Above the overload level, the BER rises owing to the distortion of the received signal.

Transimpedance (TZ) Amplifiers

High-impedance (HZ) ampliﬁers are effectively open-loop architectures, and exhibit a high gain, but a relatively low bandwidth. The frequency response is similar to that of an integrator, and thus HZ ampliﬁers require an output equalizer to extend their frequency capabilities. In contrast, the transimpedance (TZ) conﬁguration exploits resistive negative feedback, providing an inherently wider bandwidth and eliminating the need for an output equalizer. In addition, the use of negative feedback provides a relatively low input resistance and thus the architecture is less sensitive to the photodiode parameters. In a TZ ampliﬁer, the photodiode bias resistor RB can be omitted, since bias current is now supplied through the feedback resistor. In addition to wider bandwidth, TZ ampliﬁers offer a larger dynamic range because the transimpedance gain is determined by a linear feedback resistor, and not by a nonlinear open-loop ampliﬁer, as is the case for HZ ampliﬁers. The dynamic range of TZ ampliﬁers is set by the maximum voltage swing available at the ampliﬁer output, provided no integration of the received signal occurs at the front end. Since the TZ output stage is a voltage buffer, the voltage swing at the output can be increased with high current operation. The improvement in dynamic range in comparison to the HZ architecture is approximately equal to the ratio of open-loop to closed-loop gain [27]. Conclusively, the TZ conﬁguration offers the better performance compromise compared to the HZ topology, and hence this architecture is preferred in optical receiver applications.

A schematic diagram of a TZ ampliﬁer with PIN photodiode is shown in Figure 24.24. With an open- loop high-gain ampliﬁer and a feedback resistor, the closed-loop transfer function of the TZ ampliﬁer is given by

However, a trade-off between low noise and wide bandwidth exists, since the equivalent input noise current spectral density of the TZ ampliﬁer is given by

where Cin is the input capacitance of the input transistor. Increasing the value of Rf not only reduces the noise current in Eq. (24.44) but also shrinks the bandwidth in Eq. (24.43). This conﬂict may be mitigated by making A in Eq. (24.43) as large as the closed-loop stability allows [28]. However, the feedback resistance Rf cannot be increased indeﬁnitely owing to the dynamic range requirements of the ampliﬁer, since too large a feedback resistance causes the ampliﬁer to be overloaded at high signal levels. This overloading can be avoided by using automatic gain control circuitry which automatically reduces the transimpedance gain in discrete steps to keep the peak output signal constant [27].

The upper limit of Rf is set by the peak amplitude of the input signal. Since the DC transimpedance gain is approximately equal to the feedback resistance Rf , the output voltage is given by IpdRf , where Ipd is the signal photocurrent. If this output voltage exceeds the maximum voltage swing at the output, the ampliﬁer will be saturated and the output will be distorted, yielding bit errors. The minimum value of Rf is determined by the output signal level at which the performance of the receiver is degraded owing to noise and offsets. For typical ﬁber-optic communication systems, the input signal power is unknown, and may vary from just above the noise ﬂoor to a value great enough to generate 0.5 mA at the detector diode [29].

The TZ conﬁguration has some disadvantages over the HZ ampliﬁers. The power consumption is fairly high, partly owing to the broadband operation provided by the negative feedback. A propagation delay exists in the closed loop of the feedback ampliﬁer which may reduce the phase margin of the ampliﬁer and cause peaking in the frequency response. Additionally, any stray feedback capacitance Cf will further deteriorate the AC performance.

Among the three types of TZ conﬁguration in CMOS technology (CS, common-drain and common- gate TZ ampliﬁers), the common-gate conﬁguration has potentially the highest bandwidth owing to its inherently lower input resistance. Using a common-gate input conﬁguration, the resulting ampliﬁer bandwidth can be made independent of the photodiode capacitance (which is usually the limiting factor toward achieving gigahertz preampliﬁer designs). Recently, a novel common-gate TZ ampliﬁer has been demonstrated which shows superior performance compared with various other conﬁgura- tions [30,31].

Layout for HF Operation

Wide-band high-gain ampliﬁers have isolation problems irrespective of the choice of technology. Cou- pling from output to input, from the power supply rails, and from the substrate are all possible. Therefore careful layout is necessary, and special attention must be given to stray capacitance both on the integrated circuit (IC) and associated with the package [32].

Input/Output (I/O) Isolation

For stable operation, a high level of isolation between I/O is necessary. Three main factors degrade the I/O isolation [33,34]; ﬁrst, capacitive coupling between I/O signal paths through the air and through the substrate, second, feedback through the DC power supply rails and ground-line inductance, and third, the package cavity resonance, since at the cavity resonant frequency, the coupling between I/O can become very large.

In order to reduce the unwanted coupling (or to provide good isolation, typically more than 60 dB) between I/O, the I/O pads should be laid out to be diagonally opposite each other on the chip with a thin “left-to-right” geometry between the I/O. The small input signal enters on the left-hand side of the chip, while the large output signal exits on the far right-hand side. This helps to isolate the sensitive input stages from the larger signal output stages [35,36].

Using ﬁne line widths and shielding are effective techniques to reduce coupling through the air. Substrate coupling can be reduced by shielding and by using a thin and low-dielectric substrate. Akazawa et al. [33] suggest a structure for effective isolation: a coaxial-like signal-line for high shielding, and a very thin dielectric DC feed-line structure for low characteristic impedance.

Reduction of Feedback Through the Power Supply Rails

Careful attention should be given to the layout of power supply rails for stable operation and gain ﬂatness. Power lines are generally inductive, thus on-chip capacitive decoupling is necessary to reduce the high- frequency power line impedance. However, a resonance between these inductive and capacitive compo- nents may occur at frequencies as low as several hundreds of megahertz, causing a serious dip in the gain-frequency response and an upward peaking in the isolation-frequency characteristics. One way to reduce this resonance is to add a series damping resistor to the power supply line, making the Q factor of the LC (inductor-capacitor) resonance small. Additionally, the power supply line should be widened to reduce the characteristic impedance/inductance. In practice, if the characteristic impedance is as small as several ohms, the dip and peaking do not occur even without resistive termination [33].

Resonance also occurs between the IC pad capacitance (Cpad) and the bond-wire inductance (Lbond). This resonance frequency is typically above 2 GHz in miniature RF packages. Also in layout, the power supply rails of each IC chip stage should be split from the other stages to reduce the parasitic feedback (or coupling effect through wire-bonding inductance) which causes oscillation [34]. This helps to min- imise crosstalk through power supply rail. The IC is powered through several pads and each pad is individually bonded to the power supply line.

I/O Pads

The bond pads on the critical signal path (e.g., input pad and output pads) should be made as small as possible to minimize the pad-to-substrate capacitance [35]. A ﬂoating n-well placed underneath the pad will further reduce the pad since the well capacitance will appear in series with the pad capacitance. This ﬂoating well also prevents the pad metal from spiking into the substrate.

High-Frequency (HF) Ground

The best possible HF grounds to the sources of the driver devices (and hence the minimization of interstage cross talk) can be obtained by separate bonding of each source pad of the driver MOSFETs to the ground plane that is very close to the chip [36]. A typical bond wire has a self-inductance of a few nanohertz, which can cause serious peaking within the bandwidth of ampliﬁers or even instability. By using multiple bond wires in parallel, the ground line inductance can be reduced to <1 nH.

Flip-chip Connection

In noisy environments, the noise-insensitive beneﬁts of optical ﬁbers may be lost at the receiver connec- tion between the photodiode and the preampliﬁer. Therefore proper shielding or the integration of both components onto the same substrate is necessary to prevent this problem. However, proper shielding is costly, while integration restricts the design to GaAs technologies.

As an alternative, the ﬂip-chip interconnection technique using solder-bumps has been used [37,38]. Small solder bumps minimize the parasitics owing to the short interconnection lengths and avoid damages by mechanical stress. Moreover, the technique needs relatively low-temperature bonding and hence further reduces damage to the devices. Easy alignment and precise positioning of the bonding can be obtained by a self-alignment effect. Loose chip alignment is sufﬁcient because the surface tension of the molten solder during reﬂow produces precise self-alignment of the pads [34]. Solder bumps are fabricated onto the photodiode junction area to reduce parasitic inductance between the photodiode and the preampliﬁer.

Search This Blog

Integrated circuit course