Power consumption in electronic gadgets is a critical factor that involves low power circuit designs which remains a challenging and complex task for the semiconductor industry, in particular, for analog applications. Some of the basic analog circuits used in various applications are operational amplifiers, voltage reference bandgap, phase locked loops, voltage-controlled oscillators, mixers and amplifiers, analog to digital converters and digital to analog converters.
Analog technologies like BiCMOS and Bipolar, CMOS and DMOS (BCD) technologies involve a wide range of devices that are critical to analog designs. The most commonly used active devices are bipolar junction transistors (BJTs), junction field-effect transistors (JFETs), laterally diffused metal oxide semiconductors (LDMOS) and passive devices such as poly-Si resistors and capacitors. One of the critical device parameters that analog circuit designers care about in these devices is flicker noise. Flicker noise, known as 1/f noise, is mainly due to capture and release of channel carriers by the defects in the oxide. However, if the device area is small or if the number of defects in the device reduces to one or two, flicker noise discretizes itself and becomes burst or popcorn noise in BJTs. In audio amplifiers, the burst noise sounds like random shots, which are similar to the sound associated with making popcorn, hence the term “popcorn noise.” Burst noise in BJTs is a common problem in operational amplifier designs and limits their signal-to-noise ratio. They are generally observed in planar BJTs with large ß coefficients[1,2]. The burst noise is also referred to as Random Telegraph Signal Noise (RTS) in CMOS devices.
Burst Noise Characteristics
Fig. 1 shows an example of a burst noise in a PNP BJT. Under a given biasing condition, there are two distinct levels of collector current and the difference between their magnitudes is the constant given as DIC, while the switching time is given as high time τh and low time τl.
Burst noise is a two-level temporal fluctuation, which places importance on statistical distributions of their characteristics (Fig. 2). The peaks in the distribution correspond to the distinctive jumps in burst noise level. Each current level follows a Gaussian distribution, but the two levels of current together follow a bimodal distribution. The difference between the mean values of the observed peak current is characterized as amplitude variation DIC. In general, a DIC / IC variation of greater than 1% causes significant impact at circuit level. In some cases, the burst noise may have more than two different levels, signifying a multi-level burst noise.
The switching characteristics are given by “on” and “off” time of the pulse. Bursts can happen several times a second, or, in some rare cases, take minutes to occur. The high and low time constants of the signal follow a Poisson distribution given by:
These average time constants are a function of gate bias, drain bias and temperature. The gate voltage dependence allows identifying the “up” time as the capture time constant and the “down” time as the emission time. The temperature based Arrhenius plot yields the activation energy Ea for these capture and emission times. This leads to the trap level position with respect to energy level of the trap EC.
The noise power spectral density function of the burst noise has a form:
τl = Average time of pulses at low level
τh = Average time of pulses at high level
fc = Noise corner frequency
This burst noise frequency spectrum is flat before fc. After fc, the spectrum rolls-off with 20dB/decade which is also known as Lorentizan spectra. Burst noise generally occurs at frequencies at or around one kHz (Fig. 3).
Burst noise depends on the trap center location with reference to the Fermi energy level Ec. The centers which are close to 3~4 kT from the Fermi levels, generate burst noise. These trapping centers are usually the result of silicon contamination with heavy metals near the junction surfaces or lattice imperfections, such as emitter edge dislocations[1].
In the SPICE model, the burst noise is approximated by:
Where KB, AB, and fc are experimentally chosen parameters that usually vary from one device to another.
Measuring and Characterizing Burst Noise
The most common method for measuring noise is an oscilloscope. While the bias to the device is applied through SMU, oscilloscope measures the sampled output current or voltage. However, this method is cumbersome and often time consuming when one needs to measure burst noise across multi-dies and multi-wafers.
While the instrumentation for noise is generally constructed by test engineers[4], new and state-of-the-art equipment to characterize burst noise are now available in the market. A few test equipment manufacturers[5,6] have enhanced their models with burst noise measurement capability by integrating simultaneous measurements of transient currents and voltage on each channel, making burst characterization much simpler than an SMU with oscilloscope. Newer versions of the oscilloscope and the function generator capabilities have been successfully integrated with the bias measurement units, so that ultra-fast transient voltages and currents can be captured. It is to be noted that the sampling frequency of the measurement system has to be at least higher than twice of the inverse time constants to capture the burst noise. Most advanced test units are able to measure time constants in the order of 10-6 seconds down to a few seconds.
Burst Noise in Operational Amplifiers
The basic operational amplifier (op amp) has a matched input pair. Under balanced conditions, where the input voltage (Vin+ and Vin-) are the same, the output voltage is at ~Vt threshold voltage from supply. Fig. 4 shows output noise versus frequency of analog op-amps for input sweep voltage Vin-under the given test conditions. The spectral noise shows 1/f2 where a Lorentzian noise behavior is observed with fRTS ~1 KHz. Burst noise was also measured in individual devices to correlate measured circuit noise and is superimposed to the op amp signal. In this case, the scaled noise is used and is calculated based on measured gain and is given by Svo,scaled=Svo,max / (Gain)2. A good agreement between component and amplifier noise is observed, but noise is much easier to measure in the op amp, having been gained up within the wafer.
Conclusion
Burst noise can rear its head in analog products that use technologies such as BiCMOS and Bipolar, CMOS, and DMOS. It is a particularly acute concern in low-frequency, high-noise applications. Designers need to understand the causes of burst noise, which often originate because of semiconductor defects and processing issues. Armed with this knowledge and using careful design techniques, designers can mitigate the effects of burst noise and design analog circuits that combine accuracy and high precision.
References:
[1] The Industrial Electronics Handbook, Power Electronics and Motor Drives, 2nd Edition, Eds. B. M.Wilamowski, J. D. Irwin, pp. 11-1 to 11-12, ISBN: 1439802793, 2012.
[2] P. F. Lu, “Low-frequency noise in self-aligned bipolar transistors”, J. Appl. Physics, vol. 62, no. 4, pp. 1135-1139, 1987.
[3] X. L. Wu, A. van der Ziel, A. N. Birbas, and A. D. van Rheenen, “Burst-type noise mechanisms in bipolar transistors,” Solid State Electronics, vol. 32, pp. 1039-1042, 1989.
[4] Kay, Art, “Operational Amplifier Noise: Techniques and Tips for Analyzing and Reducing Noise”, Publisher: Elsevier, Newnes, ISBN 9780750685252, 2012.
[5] Agilent B1500A Publisher’s Manual, 2012.
[6] Keithley Instruments Semiconductor Characterization System Manual, 2012.