Looking back on 2007, the biggest new DSP topic has to be multicore programming. To get a sense of how much attention the topic has garnered, check out this dizzying collection of multicore programming articles published on embedded.com. Even the NY Times recently weighed in on the issue.
All this interest of course stems from a very big problem; it's incredibly difficult to program the darn things. Multicore architectures present an altogether different set of problems than traditional processors, and the programming models that have been developed to tackle them are immature and untested. As pioneering computer scientist David Patterson said in the aforementioned NY Times article, "Industry has basically thrown a Hail Mary. The whole industry is betting on parallel computing. They've thrown it, but the big problem is catching it."
The good news is there's no shortage of innovation. Numerous paradigm-shifting interprocessor communication (IPC) schemes have been developed. OSs have evolved to handle some multicore architectures. The actor programming model, once an academic endeavor, has seen application in massively parallel architectures such as Ambric's. Some companies, like TI with their DaVinci platform, have developed APIs that allow programmers to access critical functions via simple calls… the list goes on.
Surely, many of these models will fail. But some will gain traction and succeed, if only for the reason that the breakdown in Moore's law has forced hardware vendors to go multicore. At this point, there is no turning back!