Blending AI and Hardware: 3 Startups Reach for AVs and the Cloud
AI may be the buzzword of the century, with companies ranging from startups to Fortune 500 clamoring to harness the technology.
AI is much more than software and presents many opportunities and new industries for hardware/electrical engineers. For starters, EEs are the people who facilitate the deployment of AI, developing new, more efficient, and more powerful computing infrastructure. Beyond this, AI has also paved the way for entirely new fields in EE, including TinyML and LiDAR.
A general depiction of AI's proposed evolution and what could make it happen. Screenshot used courtesy of Intel
Despite being software-centric, the rise of AI has undoubtedly caused a boom in the field of hardware. With this boom, many startups have started to pop up with plans to innovate the AI industry. By looking at these startups, it is possible to see what is trending and where the industry might be heading.
Argo AI: AI for Autonomous Vehicles
The first startup that is recently making waves is Argo AI, one of the larger competitors in the race for autonomous vehicles (AVs) and the recipient of over $1 billion in funding between Ford and Volkswagen.
On the hardware front of its AI systems, Argo creates its own proprietary LiDAR sensing technology. Last week, the company released news of its LiDAR system, which it claims is the industry’s best long-range LiDAR.
At the heart of Argo LiDAR is what it is calling "Geiger-mode," a time-of-flight LiDAR that can detect a single photon.
A graphic showing the single-photon sensitivity of Argo's Geiger mode technology. Image used courtesy of Argo AI and the Ground Truth
This extremely high sensitivity is crucial for detecting low reflectivity objects, something other LiDAR can struggle with. Beyond "Geiger-mode," Argo LiDAR operates at wavelengths higher than 1400 nm. The company claims to provide them with a 360-degree FOV at a longer range and higher resolution than its competition.
Argo's technology is entirely proprietary and will probably never be available for deeper insight; however, if it's as good as it claims, more companies will likely follow in the same direction.
Kneron: AI Computer for Autonomous Vehicles
The company makes hardware for AI edge computing, with silicon designed for fields including smart homes, robotics, and autonomous vehicles. For its impressively low-power computing resources (its latest SoC offers .9 TOPS/Watt), Kneron has gained investments in the past from companies including Qualcomm and Foxconn.
Kneron’s KL720 AI SoC. Image used courtesy of Kneron
Another move besides receiving funds, Kneron has also acquired Vatics, a company known for its image signal processing SoCs. This partnership further aligns Kneron for the AV space, along with its announcement that the newly combined companies will develop full-stack AI solutions combining Vatics' signal processing IP with Kneron's edge computing IP.
With the startups progressing in AI edge computing, others are reaching towards improving on the cloud level.
NeuReality: The Next Evolution in AI Hardware
A final promising startup in the AI world is NeuReality, an Israeli startup focusing on cloud-level AI deployment.
Earlier this year, the company received over $8 million in funding for its new architecture, which brings networking to its hardware.
NeuReality is working primarily in secret. Since that's the case, many technical details behind the product are under wraps. With the bit of available information, its architecture utilizes networking techniques similar to traffic management for Ethernet packets to offload the neural networks, generating maximal performance and energy savings.
According to the company, the results are a minimal 15x improvement in performance per dollar for deep learning applications.
A high-level graphic of NeuReality's claims for their AI hardware. Image used courtesy of NeuReality
From its website, NeuReality's solution claims that it "reduces the dependency on CPUs, NICs, and PCI-switches and moves simple but critical data path functions from software to hardware. The applications of this revolutionary infrastructure include a new server class, cloud-aware virtualized SDK, AI-Server-on-a-Chip, and a disaggregated approach that enables ultra-scalability."
The company currently has working prototypes based on a Xilinx FPGA; however, the company plans to use its funding to help develop their solution on fully custom silicon within the next year.
While AI is essentially a software pursuit, it has created many opportunities for hardware engineers, with companies of all sizes receiving millions in funding to progress the industry in many different ways.
Interested in the topic of AI hardware advancements? Take a look at some recent innovations down below.