CDSI (old)

CDSI team leader: Laurent FESQUET

Event-based technologies

Event-based approach

Event-based is a quite simple idea, which suggests operating a circuit only when needed. Nevertheless, this is countercurrent when looking the semiconductor industry. Indeed, everything is clocked synchronized, analog-to-digital conversion is clocksampled. In practice, clock is used as an event generator giving the pace of the circuits generating a large number of events and producing useless activity, computation, storage or communication. The event-based approach tends generating sparse events related to natural events such as a pressure variation or a heartbeat. Therefore, the team works on alternative analog-to-digital converters able to drastically reduce the number of samples and, hence, limit useless activity and energy consumption.

Asynchronous Circuits

Since more than 20 years, the team works on new synchronization paradigms, which are not based on a clock but on handshake signals. Such techniques reveal many opportunities for rethinking the circuit design process and opening new degrees of freedom. The first expected advantage is probably the reusability of existing blocks that can simply be connected together, making the assembly of a system a kind of LEGO build. Indeed, the timing assumptions are locally fulfilled guarantying an easy block association. Moreover, many other advantages are of interest such a lower power consumption, a better robustness, lower electromagnetic emissions, safer and more secured circuits.

Targeting Ultra-low power

Today power is a main concern for chip design. The event-based strategy is probably the best technique for reducing power at least by one order of magnitude. Indeed, a sparse sampling scheme produces much less data, which are non-uniformly spaced in time. Each datum is no more than an event that can be sporadically processed by asynchronous circuits. Indeed, these latter are data-driven and consume energy only when computing. Moreover, the intrinsic robustness of asynchronous circuits favors their use at low-voltage, near- or subthreshold. Indeed, lowering the voltage is an efficient and well-known strategy to save power. Its main drawback is the decrease of the circuit speed. The Fully Depleted Silicon on Insulator (FDSOI) technology allows mitigating this speed drop thanks to forward body biasing.
As asynchronous circuits use communication protocols indicating circuit activity, the handshake signals are perfectly suited for controlling local body-bias domains ensuring low-energy expenses for body biasing and compensating the speed loss. All these mechanisms can be implemented for mitigating the energy and helping the adoption of energy harvesting in batteryless systems.

Security

Another opportunity offered by the asynchronous circuits is its ability to make more difficult the side-channel analysis and attacks in trusted devices. Indeed, the absence of clock synchronization, the specific encoding and the computation time control makes them of interest for developing trusted platforms. They also offer disruptive strategies for true random number generators (TRNG) and physically unclonable functions (PUF) while consuming a few energy.

Design flow and proven technology

Developing non-conventionally synchronized circuits is not obvious because of the lack of dedicated CAD tools. Although the first good idea is to implement such tools, there is some overcoming hurdles. The first one is clearly the quasi-absence of trained people with the know-how for designing efficient and performant asynchronous circuits. The second is the impact, the reliability and the engineer confidence into a new design flow. Therefore, for more than 10 years, the team is developing dedicated flows based on the standard commercial tools with a particular emphasis on a proven by construction synthesis.

Near-sensor computing

With the dissemination of autonomous and connected objects, it appeared the need to limit the amount of transmitted raw data, especially in RF communications where the problem is more acute. Therefore, developing tiny sensor platforms able to preprocessed data before transmitting information is becoming a challenging topic. Indeed, enhancing the sensing techniques and immediately processing the raw data with a reduced energy budget is the grail in near-sensor computing. The team developed several strategies based on event-based techniques or improving the adequacy between the algorithms and the circuit architecture. This is typically the case for many image-processing applications such as panoptic camera for laparoscopy.

Smart-sensing technologies

In-sensor computing

As previously stated, smart sensing is a key for envisioning systems with advanced features such as detection, pattern recognition or low-power. Beyond the state of art of sensor technology, the enhancement can be obtained thanks to new architectures or in-sensor computing. One of the approach concerns image sensors, which usually permanently read the image. This is a waste of energy and time for acquiring an image. In order to reduce these issues, the image capture can be performed thanks to an event-based readout, which only samples a pixel when this latter fires. In this case, the firing pixel indicates that its value has to be changed in the image memory. Such a strategy is applied for reducing the power consumption and increasing the speed sensor thanks to a dedicated readout canceling the spatial and temporal redundancies.

Measuring time

Using an event-based sensing implies a duality with the standard Nyquist analog-to-digital conversion because the quantization is no more applied to the amplitude but to the time elapsed between two successive events. Therefore, designing advanced Time-to-Digital Converters (TDC) is an important block for many sensors or even for some security primitives such as TRNG or PUF.

Harvesting for ultra-low power systems

With the advent of the Internet of Things, the system requirements in term of power are extremely demanding, especially for smart sensing and actuating. A typical highlight targets the medical implant such as pacemakers. Indeed, they need today a battery, which lasts less than 10 years. Then the pacemaker has to be explanted because this is not a rechargeable battery. In order to overcome this issue, a strategy is to harvest the heart mechanic power thanks to a piezoelectric harvester. The MEMS are particularly well-suited for extracting energy from different sources (thermal, mechanical, electrical…) for small autonomous and smart objects.

Security (chaotic approach)

MEMS have opened the doors to intense researches covering most of the technology fields. It is not surprising that they can be of interest for security. They offer original solutions for designing chaotic generators using the dynamical bistability of a Duffing’s microresonator. This approach is particularly relevant for generating true random numbers because MEMS already exist on various systems such as mobile phones and are useable for extracting chaos. Moreover, this could be employed for securing communications thanks to a couple of twin chaotic MEMS, using various transduction schemes such as electrical, acoustic or optical signals.

New Sensors and actuators

The team is also developing original micro-acoustic systems (Piezo Micromachined Ultrasonic Transducers) used as microphones, non-contact gesture recognitions, proximity sensors, fingerprint sensors or aeroacoustic measurements. Piezo-MEMS devices have been developed in order to give a haptic rendering by friction modulation. They are key components for the future haptic touch screen, which will be used in many applications (automotive, smartphones, …).