Wireless Everywhere and Programmable Designs

Robert Cravotta’s article on technology inflection points highlights the idea that electronics industry progress is often driven more by added simplification, not added complexity, in underlying technologies. When the internal sophistication of a technology can be encapsulated in robust, general-purpose abstractions, it becomes more usable.

We have seen this in the many key transitions in our industry: from analog to digital design; from hardwired to programmable systems; from custom interfaces to industry-standard I/O ports; from company-specific semiconductor fabrication design rules to foundry-based open design rules; and from processor-specific instructions sets and assembly code to C and Java programming on widely-used instruction sets

Typically these innovations are not purely technological; they often involve new business models, which allow greater collaboration, standardization, and access to the simplifying abstraction that allows wide proliferation and innovation on top of standard.

Let’s consider two important transitions in communications systems occurring right now – standardization around Long Term Evolution (LTE) wireless communication and emergence of automated processor-based design approaches.

The last ten years have seen wide proliferation of competing cellular wireless with at least three competing and largely incompatible threads: the 3GPP GSM/GPRS/EDGE/WCDMA family, the Qualcomm CDMA/EVDO family, and the Chinese TD-SCMDA family. In a notable triumph of common sense, all three standards are converging on the common 3GPP LTE standard.

A single standard means that less engineering and commercial energy will be spent in building incompatible-but-largely-equivalent designs and arguing about nuances in functionality, and more in improving the implementations and accelerating the further evolution of the standard to higher bandwidth, richer features and lower power. This simplification also raises the likelihood of more ubiquitous connectivity for all kinds of phones, net-books, consumer products and industrial widgets.

This trend will encourage embedding of LTE wireless connectivity into a huge variety of system-on-chip designs, just as wired I/O standards like USB have become nearly universal for local wired connections. We can easily picture that the hardware cost for adding full LTE capability to a platform could drop below $5. As a result, we would expect to see high demand and rapid innovation in LTE building blocks, integrated not only by traditional cell-phone chip suppliers but by every variety of volume-driven entertainment, computing and communication chip designer. We would also see increased interest in more complete design kits that allow essential LTE functionality to be extended or modified to fit the exact platform needs and cost vs. performance targets of the system.  When there’s no longer an argument about which cellular wireless standard to implement, you can implement it in everything.

The second transition is one of essential design methodology. High volume wireless designs have long relied on pipelines of hardwired function blocks implementing essential modulation, coding and protocol processing functions in transmitters and receivers.  Hardwired designs are giving way to more programmable designs because the software abstraction offers so many benefits in reducing development costs, increasing the pace for adding new differentiating features, and accommodating the combination of legacy wireless communication protocols (from those three families of rival standards) and new high-performance standards like LTE.

Steady progress in reducing processor power consumption helps enable this transition, but isn’t enough by itself. The data-rates of wireless communications is rising so much more than Moore’s Law silicon scaling, that we can’t expect faster, wider, smaller versions of older processor families to keep up. The benefits of more thorough programmability and minimum energy consumption in the baseband can’t be achieved by just putting down an array of big DSPs. Instead we see new kinds of processors that are hybrids between hardwired blocks and traditional DSPs, retaining the function unit specialization of RTL-based design, but bringing C-language on-the-fly programmability to every function subject to revision as new features requirements appear.

We also see widespread adoption of heterogeneous multi-core clusters – where a set of processors, each one energy- and performance-optimized for its specific place in the computation pipeline, works in close synchrony to implement the programmable baseband system. Most importantly, perhaps, this process of defining, selecting, clustering and programming these processors is increasingly automated. Standard GUIs, standard interconnects, standard C language coding, and tool optimization methods work together to make multi-core subsystem design exploration and hardware/software implementation as routine as embedded processor programming.

Both of these inflection points are at hand. The principal standards and technologies are now commercially available. And as with other significant inflection points, the resulting changes on downstream markets and technologies are already accelerating.

Chris Rowen, Founder and CTO Tensilica