Fig 1. Voltage references are vital components in applications like cell basestations to provide reliable levels to both the analog-to-digital converter (ADC) and the digital-to-analog converter (DAC).
Fig 2. The ISL21010, or a comparable voltage reference, provides the reference level to the ADC while the ISL28134, a very low-noise chopper stabilized amplifier, provides the input.
Fig 3. This figure shows the output voltage of the ISL21010 (a), focusing on low-frequency noise, as well as the low-frequency output noise of the ISL21009 (b).
Voltage references are one of the simplest components in any electronic system. They’re designed to provide a specific output voltage. That’s it. Batteries are powering more and more devices, and battery voltages drop as they provide power. You might trust the battery voltage to provide a crudely stable voltage level, but be careful. There are many types of battery chemistries and densities.
Lithium batteries do a decent job of holding their voltage over their lifetime, but lead acid and alkaline batteries start losing voltage as soon as they begin acting as a power source. Because of the diminishing value of a battery voltage, voltage references are used to provide a more reliable level within a system.
Although that seems simple, that reference voltage has to be guaranteed within a large array of circumstances—changing power supplies, variable load conditions, shifting temperatures, and more. Beyond the quality of the voltage reference operation in a variety of conditions, there are internal measurements of quality such as power consumption, voltage accuracy, and amount of noise. The higher the quality, the better the reference.
Consider the voltage reference for an analog-to-digital converter (ADC). Many references are available at 1.024 V, 2.048 V, and 4.096 V. These references are specially designed to serve the converter market since their values correspond to the number of millivolt levels present in 10 bits, 11 bits, and 12 bits, respectively. An accurate voltage reference ensures quality conversion between digital and analog signals (Fig. 1).
Bandgaps And FGAs
The tried and true method for producing a reference voltage is a circuit called a bandgap, which is a clever circuit based on the fundamental properties of a bipolar transistor. For example, the voltage of a pn junction, like those that can be found in a bipolar transistor, decreases as temperature increases.
The bandgap circuit uses this trait to scale a current that is inversely proportional to temperature. Then, it creates a signal that is proportional to temperature. These two signals can be combined to cancel the effects of temperature. This was big news in the late 1960s and was popularized in the early 1970s. By the way, the circuit is called a bandgap because the typical output voltage, 1.25 V, is close to the bandgap voltage of silicon, 1.22 V.
Since the bandgap circuit depends on a pn junction, you might assume that voltage references require bipolar processes. That is not true. CMOS processes have parasitic bipolar transistors that, while lower in quality, still can be coaxed into providing the necessary signals to create a bandgap circuit.
There is more than one way to build a voltage reference in a CMOS technology, though. If a parasitic bipolar bandgap is not used, the floating gate array (FGA) is an alternative. The basis for this type of voltage reference comes from the name, as it uses a floating gate. The actual gate of a CMOS transistor is left floating.
The charging process collects a certain amount of electrons on the floating gate. The number of electrons is proportional to the signal that can pass through it from drain to source. This allows for a user-programmable voltage or for the factory to set a voltage that the user needs. This voltage is incredibly stable because the electrons on the floating gate are encased in dielectric material, similar to being locked in a glass box.
Compare & Contrast
How does the FGA reference compare to the bandgap? Any design is going to have its own set of inner tradeoffs (see the table). One of the most important parameters is power dissipation. Any circuit that can burn more power can improve its performance. Therefore, I have chosen two devices with similar (within a factor of three) supply current: the ISL21009, one of Intersil’s FGA devices, and the ISL21010, one of Intersil’s bipolar references (Fig. 2).
At first glance, these two references are relatively similar. The output voltage is 2.5 V, while the initial accuracy is a little better for the bipolar. On the other hand, the FGA device can accept a much wider range of supply voltages. The output voltage and supply ranges are specifications set by the system. The other specifications will give us insight into the natural strengths and tradeoffs associated with a bipolar bandgap versus a CMOS FGA voltage reference.
Near the top of the list is the temperature coefficient. This represents how the output voltage of the reference will change with respect to temperature. The FGA reference is a factor of five better than the bipolar one. This makes sense if you remember the structure. The FGA has its charge isolated on a floating gate. The bandgap cancels the effects of two signals generated by a pn junction that is known to change with temperature. This ability to cancel may have its own characteristic performance versus temperature.
Next, let’s look at typical output noise. The FGA is more than seven times lower than the bandgap. Again, the control signals are isolated in the CMOS process instead of flowing through transistors as they are in the bipolar counterpart. This allows for a typical output noise of 4.5 µV p-p in the ISL29009 instead of the still respectable 37 µV p-p of the ISL29010.
The same trend is evident in the low-frequency noise, which is the noise measured from 0.1 Hz to 10 Hz. While the plots in Figure 3 offer similar signal sizes, note that the output noise of the FGA reference, the ISL21009, has been enlarged by a factor of 1000.
The last parameter highlighted is long-term drift. No one wants their reference to age poorly, so this number should be as small as possible to minimize changes. Again, the FGA outperforms the bandpass by a factor of two with a measurement of 50 ppm instead of 110 ppm.
Many of the FGA advantages come from the structure of the device—that floating gate. Every new solution has its advantages and disadvantages. One interesting disadvantage of the FGA topology is its ability to be affected by repeated exposure to X-rays, which can provide enough energy for the electrons trapped on the floating gate to escape their glass enclosure.
It takes quite a bit of X-ray energy to release a noticeable amount of electrons and lower the voltage on the gate. For reference, you would need to take an unshielded device through airport security about 50 times. Many times, however, these devices are mounted on the underside of a printed-circuit board (PCB) and the ground plane of the board in conjunction with other devices, and packaging can shield the FGA reference.
In some ways, the choices of example devices can be misleading. Bipolar processes tend to handle larger supply voltages than their CMOS counterparts. Bipolar circuits can be designed to exhibit lower noise and higher accuracy, if allowed a few tradeoffs like burning a little more power. And for FGA devices, the supply current tends to be quite small, making them ideal for battery-powered systems and energy harvesting. The dielectric isolation of the gate ensures that the voltage is reliable for around 100 years.
Both topologies make quality voltage references. Your system and budgetary requirements will be stronger factors in your decision. The FGA topology shows that CMOS can give the bandgap, the heavyweight in references, some solid competition. Of course, bipolar designers haven’t stopped improving their references either. New devices will continually emerge with improved performance.