There are still a lot of challenges that need ironing out for the automotive industry to move on to the next levels of autonomy.
The artificial intelligence (AI) chip industry is heating up, with huge investments being poured into a lot of startups this year. But according to Hailo CEO Orr Danon, there are not a lot of companies that are delivering a really game-changing solution that can be productized.
Danon says you’ll see many good things in PowerPoint presentations, “…but where can you order a product that actually works and gives you something that you didn’t have before?”
Many companies are showing ideas that look good—but to bring it all the way down in software and hardware and still maintain the very good levels of performance and power consumption is a very challenging task.
“It is a much more complex problem than just finding one trick that solves everything,” says Danon. “This is where you need an interdisciplinary team that can handle hardware and software design, machine learning engineering, and then bring all these together.”
Hailo is a Tel-Aviv-based AI chip company founded in 2017 by members of the Israel Defense Forces’ elite technology unit. Its first product, the Hailo-8 AI Processor for Edge Devices—which the company claims as the world’s most powerful and efficient edge AI processor—is now in production and in the hands of dozens of customers from various application fields ranging from automotive through smart cities, retail, access control, and industrial automation, to name a few.
Hailo-8 is a dataflow-based processor—wherein instead of doing all the computations sequentially, it distributes the computational flow.
“The Hailo-8 is very suitable for the processing of neural networks,” says Danon. “It’s a very unique approach, and we are the first to productize it—both from a hardware and software perspective. By using this kind of approach, we were able to cut power consumption by a factor of around 10 compared to equivalent products in the market, while still maintaining a very high level of performance.”
Many customers from different fields including the automotive, security, and industrial automation sectors find this combination very attractive because it helps them to deploy state-of-the-art algorithms into embedded, resource constrained environments, that, in the past, were only able to run on the cloud.
AI paving the way for autonomous driving
When it comes to autonomous driving, there are two critical issues, according to Danon: first is to understand what you see around you; and second is to make the decisions.
“The common denominator for all these levels of autonomy, or levels of automation, is the ability to percept, to understand the environment,” says Danon. “This will only grow with the increasing number of sensors—and the ability to process them through AI is the only game in town when you are talking about understanding and making sense of the visuals around you. This goes for anything from Level 2 to Level 5 and is also where we focus our attention.”
According to Danon, the most dominant trend is the re-emergence of the Level 2+ or Level 2++, “all these terms that people are using to define the spectrum between basic Level 2 and advances to better safety, better comfort.”
Danon says the majority of the industry is looking at surround-view vision systems and better ADAS. “What we bring to the table for these systems is the ability to process multiple sensors in high resolution and lower power budget, which today is a very big hurdle in the deployment of these systems in mass production of vehicles,” he explains. “What I do see is that most of the industry is taking a more cautious and realistic approach towards trying to best understand your environment. I think that is the right way to go in the coming years, until we establish our confidence and gain more experience with these technologies.”
There are still a lot of challenges that need ironing out for the automotive industry to move on to the next levels of autonomy. But Danon says some parts are already more mature than others.
“That’s why you see Level 2+ moving forward. Understanding the environment and having extra levels of safety and comfort are all problems that are relatively well understood,” he says.
But, for the more complex part of having things that drive on their own, and how that will look like in terms of the decisions to take in different scenarios, these things are not at the level of engineering yet, but much more in the areas of research, according to Danon.
“Although we see people starting to talk about deploying all kinds of trials in the public realm, which is a good thing and a bad thing, I assume, they are still very far from where we want to be,” he says.
What Hailo brings to the market is the ability to have the kind of performance levels that manufacturers are getting from ‘“monster” GPUs that are very good but are costly and consume a lot of power—on top of the cooling challenges. “We bring this kind of performance level in production-grade silicon devices that consume tenths of the power and cost much less,” says Danon. “We enable a setup similar to what is delivered by GPUs, but at lower power and lower cost. This is particularly important as the setup with these “monster” GPUs is only commercially viable in high-end cars which can sustain the cost and additional complexity of getting rid of the heat.”
One important decision that Hailo made contrary to some other processors in the market is not to be a closed garden, where customers are basically getting the standard package—same with everybody else—that they cannot play with too much to differentiate themselves. “Rather, we have an open platform with a great engine structure, in which each player can build their own level of differentiation according to the effort they want to invest,” says Danon.
“It’s a really exciting period for the industry,” Danon says. “Things are materializing. A year or two ago the focus was research to initial deployment. Now, things are going into real deployment and in scale. It’s a fascinating period to work in this sector.”
Apart from the “practical” applications in the automotive market, Danon is seeing opportunities also in the industrial automation sector.
“I think many companies are deploying security and management solutions for buildings or smart cities, leading to the industry growing on a much wider scale than how it used to in the past. All these things are happening and accelerating very nicely. There was some shakeup during the coronavirus pandemic, and nobody knew where things were going, but now things are back on track,” concludes Danon.
Stephen Las Marias is the editor of EE Times Asia/India and EDN Asia.