Q&A | July 18, 2014

Inventing The Stochastic Analog to Digital Convertor

Bob Verbruggen joined imec in 2005. Now he develops analogue-to-digital convertors (ADCs).

By imec


Maybe you know who invented microscopes, ballpoints, or glasses. Sometimes, inventors are well-known names that we encounter in history classes. But they may as well be geniuses forgotten by history. Companies and research institutes also have inventors. Imec’s patent group receives about 300 invention report sheets (IRS) per year. These are the documents that our researchers complete when they are convinced that they have found a unique solution. These ideas lead to over 100 patents per year. Bob Verbruggen is a regular visitor to the patent group, having written 15 invention reports in less than 10 years. We asked him to talk about his work on analogue-to-digital convertors (ADCs).

What are ADCs, and why have you chosen to work on them?

Verbruggen: You will find analogue-to-digital convertors in almost all electronic appliances. They convert the analogue signals from the environment (sounds, images, etc.) to digital signals that can be processed in a digital computer.

It was a coincidence that I became involved with ADCs. In 2005, I graduated as an engineer at the VUB (Vrije Universiteit Brussel), and I wanted to pursue a Ph.D. at imec. One of the topics available was developing ADCs for millimeter-wave applications (60GHz). At that moment, there were only very few people working on that type of ADC. So, it was a nice challenge. In addition, ADCs are very rewarding, because there are many roads to try and many possibilities for improvement. Compare ADCs to other devices, like  low-noise amplifiers for example. The latter consist of a few transistors, while ADCs have thousands.

Do you still work on ADCs for millimeter-wave applications?

Verbruggen: Imec develops ADCs for millimeter-wave applications (60GHz and 79GHz) and for reconfigurable radios. These are radios that help your smartphone to send wireless data efficiently using one of several standards (WiFi, 3G, 4G, Bluetooth, etc.). The ADCs that I currently work on are for these reconfigurable radios. The other components needed for these radios are developed by my colleagues. They include the transmitter, the receiver, and frequency synthesizers. 

 What are the biggest challenges for developing ADCs?

Verbruggen: The goal is always to develop an ADC that uses minimal energy, generates a lot of bits in a short time, is very accurate, and doesn’t take a lot of floor space. A few years ago, the main focus was on energy use. But we’ve solved that problem, and now the challenge is to get more bits and a higher speed for a given milliwat usage.

We have many techniques at hand to come to good solutions. Of course we work with the most advanced technology nodes that we have access to. When I did my Ph.D., I made ADCs in 90 nm CMOS. Now I make my designs with 28 nm transistors. Using such deeply scaled technology certainly helps to make ADCs with a higher speed.

We also work on the level of the architecture. We use flash, pipeline, and SAR ADCs. Each comes with its own advantages and disadvantages. Flash ADCs are fast but have a limited resolution. Pipeline ADCs are fast with a high resolution, and SAR ADCs are very efficient and low-power. A clear trend is that the new ADCs are made by combining various architectures.

There is also room to improve ADCs by working on the calibration. In advanced technology nodes, we can implement more complex calibration techniques, which opens up interesting new possibilities.

 Your most recent invention is a stochastic ADC. What is that?

Verbruggen: It’s a new technique that can be added to the existing architectures. It adds the number of bits without using more energy. The technique is special because it uses the statistic properties of noise in some of the building blocks to improve the accuracy. In the traditional way of working, the noise is just seen as a factor that disturbs the ADC. Contrast this to the new technique where we need a certain amount of noise. Given an ADC with noise, we rely on statistical methods to improve the accuracy.