Research is being conducted along three main axes that are very tightly coupled:
- adaptive MPSoC platforms;
- networks-on-chip; and
- model-driven engineering of embedded systems.

  
Adaptive MPSoC Platforms

We are interested in the development of MPSoC (Multiprocessor Systems-on-Chip) platforms 
that can be adapted, both at design and execution times, to the conflicting requirements and 
constraints of embedded systems that must execute application loads that vary dynamically, 
such as performance, energy, power, and real-time. Configuration and adaptive mechanisms 
may include, for instance:
- placement of cores in a network-on-chip
- dynamic or static allocation of tasks to processors
- dynamic task migration
- Dynamic Voltage Scheduling and Power Management, combined to task scheduling

Since dynamic mechanisms should not harm the system performance, we are interested in 
light-weight mechanisms, mostly based on heuristics that can find sub-optimal solutions in a 
very short time.

We are also developing a middleware for MPSoCs based on Networks-on-Chip that offers an 
adequate abstraction for programming such systems (e.g. remote invocation of tasks that are 
allocated in other nodes of the network). This middleware integrates adaptive mechanisms and 
makes them transparent to the system developer. It also allows a transparent communication 
between objects that are implemented either in software or hardware. Middleware services may 
be also implemented in hardware for performance or energy saving purposes.

 
Networks-on-Chip

We are interested in the exploration of the huge design space that is available in MPSoCs 
based on Networks-on-Chip. This design space is available due to the very large number of 
design decisions to be taken in such systems – placement of cores and memory nodes, task 
allocation, memory organization, router architecture, etc. Design solutions must consider 
conflicting requirements – performance, energy, power, real-time, thermal constraints.

Experiments are being conducted upon a SystemC virtual platform built on top of the SoCIN 
network-on-chip, developed at UFRGS.

  
Model-Driven Engineering of Embedded Systems

The architecture of embedded systems combines hardware and software components, to 
which different application functions must be mapped. The architecture is usually designed 
as an instance (a configuration) of a pre-defined platform. Finding the right architecture and 
the right match between application functions and architecture componentes is a hard design 
task, due to the very large design space to be explored.

We are interested in using the Model-Driven Engineering (MDE) approach, using application 
and platform models that follow appropriate meta-models. Model transformations are thus 
applied to find the right mapping between application and platform and to generate an 
adequate implementation. UML is used as language for expressing all those models. The 
mapping task is performed in a design space exploration context, where adequate heuristics 
(such as simulated annealing and ant colony optimization) are used to find solutions that 
meet design requirements and constraints.