The global message I took back from ARM TechCon is that companies like ARM, Synopsys, Tensilica (a part of Cadence), and other IP MCU vendors have a difficult task ahead of them. None of them develop leaf nodes in the IoT market, yet they must anticipate the demands that third party architectures put on their products.
During the conference I focused on three product areas: IoT, Automotive, and Security.
The term IoT is vague since Internet of Things has so many meanings and covers so many architectures to make a precise discussion of the term meaningless. In general a IoT architecture implementation has three functional levels: leaf nodes, gateway, and cloud computing. Although the nature of a leaf node varies to the point that any market can be served by one of them, its functionality has been accepted by most. A leaf node acquires data through sensors, and provides stimuli that are converted in responses to the data collected. In addition a leaf node provides inputs to the gateway node and may react to data from the gateway.
It is also practically universally accepted that the cloud computing part of IoT serves as a data mining operation that produces global information about the entire specific IoT system. The data analysis provides input to the gateway and from there to leaf nodes in order to achieve the goals set for the system.
There is no generally accepted definition of a gateway. Although its function is to aggregate inputs from a varied number of leaf nodes and perform some analysis,and to provide control inputs to the leaf nodes, its nature is not defined. A gateway could be a smart phone, a home gateway, or a special function box. Its nature depends on what the IoT system is meant to do. The problem arises when a vendor is designing a smart phone, for example. Adding functionality to support IoT functions will increase its cost and may make the device uncompetitive, yet leaving the functionality out might generate the same result. Will we need to have different versions of the same phone? And if so, do we need different versions of the MCU used in the phone?
Home gateways now provide communication functions for computer systems, Television, and other entertainment systems, but they lack the local computing power and security apparatus to be an appropriate IoT gateway. New devices will be required, what architectures and what computing capabilities will satisfy the requirements?
The intelligent car and the self-driving car use so many different electronic devices that one or more IoT systems are implemented in a car. Motor Control, Advanced Driver Assistance System (ADAS), Infotainment are all examples of IoT architecture that in one form or another are present in new cars. The importance of electronic systems in the car has grown to the point that ARM has created a new division under the leadership of Lakshmi Mandyam to serve the market. The immediate attention of the division will be: algorithm complexity, motor control, ADAS, and vision processing. Instead of “algorithm complexity” you can read artificial intelligence algorithms.
But the future does not just rest within the car itself. Vehicle to vehicle (V2V) communication and the intelligent road are also components that will make the self-driving car a reality. Cars need to communicate with each other and have a way to resolve conflicts with respect to the next moves the two or more cars will execute. At a signaled intersection, the signal must provide electronic signals to manifest the status of the flow of traffic. And what about emergency vehicles or stop signs? I have found a few stop signs hard to see due to foliage or other obstructions that were not present when the sign was originally installed. Will a camera be able to scan for signs that are at different height, partly obscured, faded, or otherwise not as obvious as a new sign?
Much of ARM presentations had to do with security. Security is another characteristic of IoT that is not well implemented and standards in this area are lacking. Simon Seagars, ARM’s CEO went so far as coining Insecurity of Things term to describe what IoT meant. Vendors serving the IoT market are developing solutions, but they are mostly covering local problems, not system wide security.
ARM has developed the Platform Security Architecture (PSA) whose goal is to separate trusted applications from untrusted ones. The execution environment is thus divided into two environments. One contains trusted data and functions, while the other can execute applications that have no connection with the trusted environment.
Synopsys has a similar approach in its ARC family of products. In both cases the local execution environment is much more robust than before, but untrusted data can still be developed locally and transmitted throughout the system.
The result is that one cannot describe the IoT system as secure unless all of the nodes, including gateways and the cloud, have the same or equivalent architecture. In this case the system is secure, but transmission of infected data to other systems can still occur. It is clear that security is not just a local problem, it is an internet wide problem and I am not sure that full security can be imposed to the internet as a post-design effort. The internet was designed to provide communication among trusted friends. Malicious use of the net happened later and thus the insertion of security mechanisms in the protocol is not easy.
Some Other Observations
I am told that the exhibit floor was smaller than the previous year. I saw many more EDA tools vendors than what I expected and fewer leaf node vendors than what I expected. Did ARM’s customers not think that the conference was a good place to advertise their products?
Much was said by ARM’s executives about the total strategic independence from Softbank, the owner. But it is really too early to tell. ARM must produce the expected revenue and profits or things will change. I am sure that Softbank did not purchase ARM for glory or sport. In fact Softbank expects that ARM’s product will find significant use internally the conglomerate, and internal sales usually have lower margins.
But there is enthusiasm and self-belief among ARM’s executives, so the near future is promising or, at worst, stable. With such large installed base ARM will be around for quite a few years.