The ECU diet: heterogeneous computing could save power, weight and space in the vehicle of the future

While heterogeneous computing could help automakers to produce better and lighter electric and autonomous vehicles, it will require a change of mindset. By Jack Hunsley

Likening a connected vehicle to a computer on wheels has become a common analogy. With more connected features and autonomous safety systems, more computing power is being placed in the hands of the driving system. However, for an industry in which IT has never been a core competency, this transition raises interesting questions for the impact on vehicle architecture.

New vehicles boast more computing power than ever, which has created a need for more electronic control units (ECUs) – in some cases upwards of 100 individual ECUs. However, there are already concerns that this architecture model may not be sustainable. “The number of ECUs just keeps rising,” explained Marques McCammon, Vice President at Wind River. “If we continue to just increase the number of ECUs in a car every time we want to add a new feature, at some point in time we will run out of space.”

This is in the crux of the problem. While at the moment the industry is still dominated by gasoline and diesel powertrains, once electrification becomes the norm both power and physical space in the vehicle will become precious commodities. Weight is also a significant concern; according to analysis from Frost & Sullivan in July 2018, the powertrain of a battery electric vehicle (BEV), with a 35.8kWh battery pack and a 100kW electric motor, can weigh nearly 125% more than a standard ICE vehicle powertrain, limiting battery range and efficiency. For autonomous vehicles (AVs), a 2018 report from the University of Michigan found that the added weight of sensing and computing subsystems limited powertrain efficiency and the amount of useable space in the vehicle. The study also found that such autonomous technology could increase energy usage and greenhouse gas emissions by between 3% to 20%.

At Wind River the solution is believed to be simple: reduce the number of ECUs through the use of heterogeneous computing. However, making the transition will require a complete change of mindset in automotive. “Heterogeneous computing is not something that we as an industry typically do,” said McCammon. “At the moment we only do it to align functions that are very similar to each other. What we need to do is combine functions that have absolutely no relationship: functions that have completely different methods of maturing and validation, timing cycles and value propositions, and bring them together in one space to work in harmony.”

If we continue to just increase the number of ECUs in a car every time we want to add a new feature, at some point in time we will run out of space

This need for harmony is vital, especially with the industry rapidly advancing towards autonomous transportation. For example, while any bugs that may occur when combining an infotainment unit with a radio may be harmless, the occurrence of a similar error when combining these systems with safety-critical autonomous technology could be catastrophic.

“A window controller and seat controller could run on the same control module because they’re similar in what they do and how they function,” explained McCammon. “That makes some sense, but the notion of having a braking system and a window system on the same controller is foreign. Those are the kinds of challenges we’re looking at.”


Making this transition to heterogeneous computing is not a jump which can be made blindly. Therefore, in order to begin to tackle the problem Wind River has come up with a four-step process known as ACRUE (Abstraction, Consolidation, Reuse, Extrapolate). “In the past, every piece of software that we brought into the car was intimately tied to the hardware upon which it sat and the applications that ran above it,” explained McCammon. “The first thing to do is to break those bonds through abstraction. After that, we can consolidate, and use virtualisation to have every function in the car believe that it’s in control when in fact it’s sitting inside of a box that’s catered and built especially for it.”

The next stage – reuse – will then see older software recycled when new iterations are released. “The amount of software that we carry over is quite limited,” McCammon continued. “If we could get to a point where we reuse a considerable amount of that software, the same way that automakers reuse much of the content in a car, we can drive down cost and improve return on investment.”

We’re bringing software into the automobile and it brings with it a consumer electronics mindset

Finally, the ‘extrapolate’ stage would require an adoption of a consumer electronics mindset in regard to product lifecycle. “We need to look at other domains where software is more prevalent or has been a critical element of a value proposition for longer periods of time,” said McCammon. “In the consumer electronics space, everyone has been on that bandwagon for some time but the reality is that consumer electronics does not have the same rigours that automotive does. We need to look at other places where the critical nature of software delivery is a little bit higher.”

Change of pace

Other industries such as aerospace made the switch to heterogeneous computing some 20 years ago, according to McCammon. Automotive, however, is lagging behind. “In the automotive industry we are very controlled and deliberate,” explained McCammon. “We do not typically release anything until we have proven, without a shadow of a doubt, that it has no flaws or weaknesses. But as we’re putting more computing power inside the car, the automotive industry’s and the software industry’s pace of product development has become grossly dissimilar. These two cultures are clashing.”

This evolving mindset revolves around the different expectations of the consumer electronics market compared to automotive. For example, while it can be reasonably expected that a new version of any major smartphone will be released on a yearly basis, the lifecycle of a vehicle model can be as much as five to seven years.

This discrepancy in lifecycle is limiting the automotive industry’s ability to introduce heterogeneous computing. While a connected vehicle or an AV can be refreshed using over-the-air (OTA) updates, the overwhelming majority of vehicles on the road cannot be altered remotely after they leave the showroom and instead require specialist assistance at garages and workshops.

New value

While the benefits have been felt elsewhere, this is not to say automotive should instantly make the switch to a more consumer electronics friendly life cycle. Concerns, for example, revolve around an adjustment to how automakers would earn money from a vehicle. What many are beginning to realise is that shared, connected and autonomous services are set to change the current value model of a new vehicle.

While the pace of change that we see now is much faster than anything that automotive has seen in its history, by everyone else’s standards it is still slow

“Consider how the automotive supply chain works: one of the first thing automakers do when building a new model is contact their suppliers,” explained McCammon. “They will tell the suppliers what they need, the suppliers will provide a product, and the product goes in the car. If these products have connected features that interest consumers, then the cost to the automaker is going to increase as the suppliers can charge more. For automakers that is not sustainable.”

What is needed therefore is an adaptation of the industry’s mindset. If, as some analysts predict, around 50% of a future vehicle’s market value will come from software, automakers need to take control of how this software is delivered to them. By asserting control and leveraging competition, automakers could manage their bill of materials in a more cost-effective manner. “What enables a healthy industry is the ability to create competition amongst its constituents,” said McCammon. “That’s not to say that it has to be dog-eat-dog; there are some places where partnerships and standards will arise.”

In short, change is coming. The value of making the switch to heterogeneous computing, especially in regard to power, space and weight savings, will be extremely valuable as the industry moves closer to EVs. However, this change will not come overnight. “Automotive is slow to change,” concluded McCammon. “While the pace of change that we see now is much faster than anything the automotive industry has seen in its history, by everyone else’s standards it is still slow. If we want the industry to be as healthy and vibrant as it has been in the past, then at some point it is going to have to transform itself.”

This article appeared in the Q1 2019 issue of M:bility | Magazine. Follow this link to download the full issue.