Heterogeneous integrated systems are based on devices that do not necessarily scale according to "Moore's Law" and provide functional diversification i.e. additional value in different ways, such as power generation and management, or sensing and actuating with motion sensors (accelerometers, gyroscopes and magnetometers), environment sensors (light level, color, temperature, pressure, humidity), or biosensors (eNoses, blood pressure, pH). The "More-than-Moore" (MtM) approach allows for such non-digital functionalities to migrate from the system board-level into the package (SiP) or onto the chip (SoC). The number of heterogeneous functions is growing at a rate of 1.3x per year in consumer portable applications such as smartphones, and opportunities are growing for heterogeneous component integration with the advent of technologies such as 2.5D and monolithic 3D. It is anticipated that the market of analog and heterogeneous ICs will experience an explosion in the 2020s, due to the strong demand of billions of connected objects with intelligent sensors, in the context of Big Data and IoT. Future integrated "sensor hubs" will require the simultaneous analysis and fusion of data from a wide range of sensor types for full interpretation of device state and situation.
However, while MtM process and device technology have progressed, the development of enabling design technologies has not kept pace. Some of the many design technology challenges are: heterogeneous system partitioning and simulation; analog and mixed-signal design technologies for sensors and actuators; new methods and tools for co-design and co-simulation of SIP, MEMS, and biotechnology. Without radically new design technology, it will be impossible to realize the full potential of MtM technology.
This research track contributes to the efficient design of MtM ICs by providing innovative design techniques and tools for the multi-physics parts of these systems. The complexity of this task, with many inter-related degrees of freedom and performance metrics, is similar to searching for the center of a very large three-dimensional maze with a flashlight and no map. The work targets the solving of fundamental issues with the way in which heterogeneous components are designed and integrated into systems during the design process. The objective is to examine ways in which to predictively extract and exploit system and circuit design information, and apply it in the context of a multi-physics sensor hub application.
Highlights of recent and current work include:
the implementation of a meta-optimization strategy for early, deterministic and optimal floorplanning of 3D ICs taking into account planar and vertical interconnect resources, inter-core network traffic, thermal variations and manufacturing cost;
the development and patenting of a predictive method for the synthesis of multi-physics systems based on a hierarchical approximation and combination of sub-system Pareto fronts. Tools associated with this method are the focus of a laboratory spinoff project;
the proposal of a new method to estimate the bounds of performance variability in analog circuits, combining design of experiments techniques (to extract and model performance variability) with the Cornish-Fisher expansion (to approximate the quantile distribution);
the efficient extraction of predictive models based on Pareto-fronts and a fast graph-based approach, and their extension to n-performance design spaces using the Hyper-Space Diagonal Counting (HSDC) methodology
This work is the result of 9 PhD theses (1 ongoing) and 3 postdocs, and has been carried out in the framework of several projects under my scientific coordination (regional program Osmose, Auto-BAW, Centre Jacques Cartier projects, CNRS PICS, Nano2008 / Nano2012 agreements, ANR-PNANO 3D-IDEAS, ANR-AAPG HELICITY, PEPR Electronique AC3 CHOOSE) as well as Eureka projects CATRENE 3DIM3 and CATRENE H-Inception).
Collaborations: CEA-LETI (FR), STMicroelectronics (FR), iCube (FR), IMEP-LAHC (FR), LCIS (FR), Asygn (FR), Intento Design (FR), TIMA (FR), TU Delft (NL), Fraunhofer (DE), Ecole Polytechnique de Montréal (CA).