Integrated circuits : Controlling the design is a major issue
Two large-scale events celebrated the achievements and potential of high tech and digital technology in June: Vivatech, where the Institut Polytechnique de Paris and its start-ups were represented in full force, and the World Mobile Congress in Barcelona.
Yet, at the same time, a global shortage of semiconductors was affecting industries around the world, bringing many production lines to a halt and focusing the attention of the international media on the near monopoly of a few Asian countries, led by Taiwan and South Korea, on this sector that had become as strategic as oil production was in the previous century.
The Covid-19 pandemic has accelerated the structural trend of increasing demand for integrated circuits.
In a world that is increasingly interconnected, more automated and more concerned about climate issues, economic growth is becoming more and more semiconductor intensive.
Their crucial role for prosperity but also for national sovereignty, coupled with the growing rivalry between the United States and China, is likely to lead to a redistribution of the cards.
The American strategy - defensive with the objective of relocating production units and offensive with the declared desire to control intellectual property - may lead to long-term domination by the United States and its allies of the semiconductor industry. China, which is accelerating the development of its production capacities, is struggling to move upmarket. In the short to medium term, however, Asian producers should continue to benefit from this structural evolution in demand and find themselves at the center of its geostrategic consequences.
While ASICs for particle physics may seem far removed from these issues, Christophe de La Taille, director of the OMEGA Laboratory (a joint CNRS/IN2P3-Ecole Polytechnique unit), points out that this field of research is still connected to them.
The OMEGA laboratory designs ASICS for particle physics. Why should these ASICs be designed in a dedicated laboratory?
Christophe de La Taille : There are three main reasons why the OMEGA laboratory (a joint CNRS/IN2P3-École Polytechnique unit) designs its own integrated circuits or ASICS (Application Specific Integrated Circuit). In the field of particle physics, we must be able to detect the presence of particles that are sometimes elusive and that are the subject of very large-scale international experiments often mobilizing several hundred or even thousands of researchers, such as those that led to the discovery of the Higgs Boson at CERN.
We therefore have very particular needs for joint development and even optimization of detectors and integrated circuits that will allow the conversion of the detected signals. Advances in ASICs allow the development of new detectors and advances in detectors require new developments in ASICS.
A particularity of our ASICs linked to this coupling with the detectors is that they include a rather important analog part. Today, 90% of the semiconductor market corresponds to digital chips that process bits. Our detectors receive analog signals, very weak signals that have to be converted into bits so that the information can be used, and so our ASICs retain a significant proportion of analog technolgies, roughly speaking our ASICs are half analog, half digital. The most common chips used in consumer electronic devices - computers, mobile phones, DVD players - are digital chips and are produced in large quantities, although there may be shortages.
We also have the very specific constraints of high-energy physics, which are problems of resistance to radiation in very severe environments of the order of hundreds of mega rad. These problems were first addressed by developing radiation-hardened technologies, but the miniaturization of transistors, a factor of performance improvement, also favored radiation resistance, which made these technologies less necessary. On the other hand, radiation is the cause of another type of problem - single event effects - which has become more pronounced with miniaturization and affects digital circuits.
When a particle has enough energy to deposit in a transistor it can change the value of a bit from 0 to 1. While these changes can be benign, they can also be destructive. If it is the self-destruction bit of the rocket that is affected...it can be disturbing. The treatment of these problems involves the architecture of the circuit, which becomes more and more complex as the radiation rates are higher, knowing that the smaller the technologies, the lower the load required to modify the value of a bit.
Finally, as circuits get smaller and more complex, we are faced with increasing problems of neighborhood between blocks that may interact when they should not or situations where very weak signals that we would like to be able to pick up are disturbed by noise generated by other blocks in the circuit. These problems did not arise when systems were much larger. The race to miniaturization may therefore be more of a problem than a solution for us. It concerns completely digital circuits, which are not ours, and may even lead to a reduction in analog performance.
What are the consequences of these evolutions on the design of ASICs?
Our laboratory does not do research on microelectronics as such, we do not try to make new transistors smaller, more this or more that, we use existing transistors which evolve regularly. We focus on the design of integrated circuits for very specific uses and environments.
We start with the basic materials available on the market, but there is a very creative dimension in designing new architectures to address the various problems I mentioned and to improve performance and adapt to the constant changes in technology.
The increasing complexity of the problems we face means that design teams are becoming increasingly specialized, which requires a minimum critical size. Twenty years ago, ASICs were much simpler, there was less international competition and one or two people could work on a circuit.
Today, the increasing complexity leads to specialization within the teams, which must also be able to work on several projects and thus be able to reuse certain blocks or architectures that they have produced or exchange them for developments carried out in other laboratories in the context of international projects.
What are the links between ASICS designers and the semiconductor industry and what are the stakes in terms of sovereignty?
We do indeed design our circuits, we test them but we do not manufacture them. We turn to founders who are now mainly in Asia, in Taiwan, South Korea and Japan. It is the Taiwanese TSMC that produces our ASICS.
Europe and even the United States have lost this production battle, perhaps because we did not really fight it. Discussions are underway in Europe and the United States on the relocation of semiconductor production, which is a sovereignty issue.
Mastering the design of advanced integrated circuits remains a major issue of sovereignty. They are found in strategic areas at the junctionbetween the civilian and military sectors, such as space exploration and satellites.
What technology transfers have resulted from ASICS' particle physics design work?
The most classic example of technology transfer is in medical imaging. The two fields are very complementary to the extent that some particle physics and medical imaging conferences are common.
The image sensors developed to follow the trace of particles have allowed the rise of CMOS sensors that equip all digital cameras and has developed into a major industry with some founders specialized on imagers.
The developments that we and other design teams around the world are carrying out in terms of radiation resistance, very precise time measurement, very weak signal measurement, etc. end up having industrial or societal applications.
The lidars that will equip the autonomous cars of tomorrow are another example of a very precise distance measurement application. The first lidar was designed to make very precise measurements in atmospheric space.
Has the work of your laboratory been disturbed by the bottlenecks that have disrupted global chip production in recent months?
The bottlenecks that have disturbed the global chip industry have only impacted us very marginally. Our production needs are rather small. The entire production of ASICs for particle physics since its inception represents perhaps only one week's production of a major chipmaker like TSMC. On the other hand, we have other problems which are related both to the qualification of the foundries we work with and to the costs induced in particular by the effects of miniaturization.
The development process of an ASIC in our domains is spread over several years between the design itself, the tests, the selection of the production lines which will allow to produce them in good conditions of reliability, the prototyping... the process takes from 5 to 10 years and can involve several actors. We therefore need to have partners with whom we can work over time and who will have production lines that will remain in operation until the production phase.
Another concern is production costs. Each time we change the technological node, the price doubles. When we manufacture 12 wafers of our prototypes in 130 nanometer technology, it costs us around 250,000 euros, when we switch to 65 nanometers it's 500.000 euros, in 28 nanometers it is more than a million euros. Fortunately, 130-nanometer production lines still exist, and even 250-nanometer or 350-nanometer, and they produce for industrial customers, particularly in the automotive sector, which helps us... but miniaturization also has a disadvantage for us in terms of cost without always improving performance.