I wrote earlier that the sheer vastness and potential for IoT designs require a different way
of thinking about system implementation. There just isn't a single controller architecture that meets
all IoT requirements, such as multiple sensors, an innovative user experience, wireless
protocol support, and various security requirements.
And if a single controller architecture is not the answer, do we just lean on other traditional processor architectures and approaches? We can't because they aren't necessarily broadly applicable in the IoT. Consider these traditional approaches to SoC design:
- CPUs: They are generally good at handling unpredictable and varied software streams, but not particularly efficient at given data-intensive tasks.
- Hardwired RTL: This approach is excellent for specific tasks and hardware optimization and customization but offers little in the way of future-proofing and flexibility once the RTL is locked down.
New Vision: The IoT Architecture
Since IoT is such a profoundly new and diverse application space, we need to consider a new architectural approach—an IoT architecture.
IoT and natural user interface applications are going to require computing at a complexity level far above today's requirements. This will push our technology even more to provide these capabilities at even lower power and with foolproof security.
Companies designing toward these applications—because of cost, time to market, design productivity, and flexibility considerations—are going to demand:
- Access to a broad design and verification IP portfolio
- A design methodology
- Analog and digital expertise
- Low-power expertise
- Security solutions
- Understanding of expanding sensor capabilities
- Low-power digital signal processors (DSPs)
- Processor design flexibility
This is the superstructure of the IoT architecture. It is an approach to processing that fits hand-in-glove with the demands of the new application space and is uniquely sensitive to the strict power limitations inherent in these applications.
Today's strategy of offloading data-intensive tasks to a specialized DSP will expand quickly in this new space. To meet this demand, it will be essential to be able to quickly generate new processors and DSPs to efficiently implement new algorithms and ensure hardware-level security.
These can't be created by hand; there's no time, and the cost would be crushing. Teams need to leverage a processor generator that also creates matching software tools.
This new approach minimizes footprint and maximizes performance while keeping power in check; it realizes that IoT is—at its core—an energy-driven application, where operations per watt is a crucial metric.
Perhaps most importantly, it's insurance: You can't design an IoT SoC today without future proofing it in a way that ensures you will support additional algorithms that you want to run for future solutions.
Coming Together
What does this mean for the success of natural user interface design and, more generally, IoT applications?
The design approach doesn't live in a vacuum but serves as the launch pad for application optimization across a broad swath of uses, such as vision, voice control, gesture, and object recognition. Think about modules for sensor fusion and sophisticated natural UI processing, low-power analog blocks and peripheral IPs, and various communication protocols such as Bluetooth and WiFi.
When combined with these other IP building blocks, compatible IoT-application-specific modules begin to take form, allowing designers to then work with these subsystems for ease-of-design or time-to-market considerations as they attack new applications.
Consider this in the context of the evolution of a popular class of IoT wearables: Fitness gear (Fitbit) and smart watches (Pebble).
A block- and subsystem-based approach to future designs might enable voice command and gesture recognition in a power-, real estate- and cost-efficient manner for a class of devices that's obviously keyboard free and very likely touch-screen free, given size constraints on the wrist.
These applications are also going to require unprecedented integration. Current versions of the Fitbit include five major ICs. An IoT architecture approach might allow customers to develop a single-chip solution. This move from five chips to one would lower cost and size, and increase battery life.
Seow Yin Lim
Related Stories:
- IoT Focus: Natural User Interface Design Crucial to Success
- IoT Focus: Wrestling with the Design, Time to Market, and Cost Challenges of IoT