Appendix A

Reference Voltage Generator

A.1 Introduction

In any A/D converters, some reference voltages are generally required to set a reference for the sampled input to be compared to. The accuracy of the reference voltages need to be as linear as the converter itself in most cases. For example, in flash converters, reference voltages are compared with the sampled input signal. Any error present on reference voltages will be added directly to the nonlinearity of the converter. The problem becomes even more severe at high resolution and high speed. In a high speed converter, switching noise on the chip can be coupled onto the reference lines and corrupt the conversion process.

Traditionally, there are two ways to generate reference voltages either by using a resistor string or capacitor array. Each one has its own limitations. It will be shown in Chapter 3 that with some architectural differences, the number of required voltage references has been reduced to two and the tolerance is relaxed with some trimming capacitor.

A.2 Resistor String Voltage Reference Generator

By using multiple passive resistors, one can generate several potential between supply and ground. In the case of a flash converter, the number of reference voltages required is , where N is the resolution of the converter. By using equal value resistors in a string, one can interpolate different reference potentials between the supply and ground (or full scale input). Two major problems seen at this point are the matching of resistor and high speed operation.

Since the flash converter relies on the absolute value of the reference voltages, mismatch between the resistors during process will cause nonlinearity on the reference voltages. This directly affects the linearity of the output bits. For high resolution, where the LSB size is small, the tolerence on voltage references is even tighter.

For high speed operation, many sampling capacitors might be switched to a reference voltage on the resistor string. This will create a glitch on the refernece, and needs to settle (with the RC time constant) to the required accuracy within the period allowed. The largest RC time constant appears in the center tap of the resistor string, where the equivalent resistor value is

Assuming high power can be tolerated, the mismatch of resistors in the process will still determine the overall linearity of the A/D.

A.3 Capacitor Array Reference Voltage Generator

Another way to generate reference voltages is to use an array of binary weighted capacitors. A family of A/D converters based on this idea is called the successive approximation ADC's. The input is first sampled onto capacitors and then compared with a reference voltage to determine MSB. Then, the quantized MSB is added or subtracted from the input signal to zoom in to next bit resolution.

The typical binary weighted capacitor array is shown in Figure 2

Although this method does not require static power, the accuracy of the reference generation still relies on the absolute matching of the capacitors. Special techniques have been introduced over the years to improve the matching, however, without special trimming, the achievable resolution is about 8-9 bits.

It will be introduced in later chapters, with some architectural differences, one can eliminate multiple reference voltages and the dependence on the absolute capacitor matching.