Smart products are already disrupting and transforming industries, with growing demand from both consumers and businesses. The smart home market alone, just one piece of the smart product spectrum, is projected to be worth $135B by 2035.1
There are several reasons for this trend toward "smart everything." Smart products offer natural interaction: many products, from home entertainment to automotive interiors to industrial equipment, can be controlled by voice and gesture rather than physical buttons or control panels. Today’s smart products let people focus on other activities by operating with increased autonomy, like robot vacuums that clean the house or the aisles of the grocery store. They enable data collection and analysis, serving everything from monitoring our health to predicting when equipment needs to be maintained.
See neuromorphic computing in action
We collaborate with external researchers in both industry and academia to push the boundaries of technology. Watch the following examples of recent neuromorphic computing projects we have worked on.
Scientific understanding of how the brain works is not yet complete, but it is mature enough to uncover many core principles of neural computation. Researchers and engineers have worked together to develop algorithms and processors that replicate some of those core principles and mechanisms.
What are they trying to emulate? An average human brain contains 80 to 100 billion neurons that are each highly efficient. Activity in the whole brain is much sparser than traditional computer architectures. Complex sequences of spikes in organic nerve fibers are nothing like the 64-bit silicon data buses we see in general-purpose processors. In the brain, each neuron works asynchronously to provide massive parallelism—many different processes all happen at once—and to adapt quickly to rapid changes in the environment.
We’ve seen a lot of progress in scaling and industrialization of neuromorphic architectures. Still, building and deploying complete neuromorphic solutions will require overcoming some additional challenges.
Neuromorphic systems are several orders of magnitude more energy efficient than general purpose computing architectures.
Neuromorphic systems excel at processing continuous streams of data and deploying neuromorphic processors at the edge reduces the delay to analysis.
Neuromorphic system architectures let devices adapt to changes in context.
Recent advances in training neuromorphic systems have enabled rapid learning from little data—capabilities beyond most conventional AI systems.
Every organization needs to shape its computational variety strategy to meet growing demands from consumers—and to stay ahead of increasing competition. Now, with emerging neuromorphic hardware and maturing platforms, it’s time to start experimenting with neuromorphic computing, starting with applications that require efficient, responsive, and adaptive AI at the edge.