Major stories this week: developers conferences, autonomous driving, neural networks, HEV/EV

Developers conferences from three of the major global IT heavyweights took place in a mere ten-day timeframe: Facebook “F8” from April 30th to May 1st in San Jose; “Microsoft Build” from May 6th to May 8th in Seattle; and “Google I/O” from May 7th to May 9th in Mountain View. Each of these events brought a plethora of announcements, and – as distant as they may seem from chip design and manufacturing – some of them will inevitably have an impact well beyond the developers’ communities of these three companies.

Early May announcements overload

Just a brief, non-exhaustive summary. Pledging to “a privacy-focused social platform”, Facebook introduced many new functions across all its apps and products, including Instagram, WhatsApp, Messenger and AR/VR. The Facebook app itself will get a new design, new group-oriented features, and a match-making function called Secret Crush aimed at exploring relationship opportunities within a circle of friends – without risking an embarrassing moment: the two persons will be notified only if both express interest for each other. WhatsApp news include the possibility for people to see a business catalog right within the app when chatting with a business; Instagram users will be able to simply tap on their smartphone screen to know exactly what ‘creators’ are wearing and buy the same outfit on the spot; “Portal from Facebook” and Portal+ will expand in Europe this fall; and Facebook’s two newest virtual reality headsets — Oculus Quest and Oculus Rift S — will start shipping May 21.

Mark Zuckerberg at F8. Image credit: Facebook

Microsoft announcements include, for Microsoft 365, the general availability of Microsoft Graph data connect; Fluid Framework, a new web-based platform and componentized document model for shared interactive experiences; the new version of Microsoft Edge; Kubernetes Event–driven Autoscaling (KEDA), an open source component that supports deployment of serverless event-driven containers on Kubernetes created in collaboration with Red Hat; and the open sourcing of the Q# compilers and simulators for quantum programming.

Microsoft Build logo. Image credit: Microsoft

Google announced that Google Lens will be able to read out loud the words at which you point your smartphone camera – such as street signs or ATM instructions – or scan a restaurant menu to highlight popular dishes and show reviews from other customers; in addition, the Duplex technology will be extended to allow Google Assistant to book rental cars and movie tickets on your behalf. The company also introduced the new smartphone Pixel 3a and the smart home device Google Nest Hub Max. Announcements also concern Google’s machine learning “ML Kit”, with the addition of three capabilities in beta: On-device Translation API, Object Detection & Tracking API, and AutoML Vision Edge. Last but not least, Google Cloud announced that its scalable supercomputers for machine learning, Cloud TPU Pods, are now publicly available in beta. A single Cloud TPU Pod can include more than 1,000 individual TPU chips (tensor processing units).

The Google I/O event 2019 at the Shoreline Amphitheatre in Mountain View. Image credit: Google

Pruning neural networks before training

Even AI experts from Facebook, Microsoft and Google are probably asking themselves the following question: since neural network pruning can reduce parameter-counts by more than 90% without harming accuracy, why don’t we train these smaller architectures instead of the whole networks? Reason: the architectures uncovered by pruning are harder to train from the start. But MIT CSAIL (Computer Science and Artificial Intelligence Lab) researchers recently reiterated that a standard pruning technique naturally uncovers subnetworks whose initializations makes them capable of training effectively. Based on these results, last year they articulated the “lottery ticket hypothesis”: dense, randomly-initialized, feed-forward networks contain subnetworks (winning tickets) that – when trained in isolation – reach test accuracy comparable to the original network in a similar number of iterations. They presented an algorithm to identify winning tickets and a series of experiments that support the lottery ticket hypothesis.

More autonomous driving tests

Trusting neural network vision capabilities, more governments worldwide are authorizing autonomous driving tests in real traffic situation. Among them the Italian government, that has recently granted the first authorization of this kind to VisLab, a spin-off of the Parma University that in 2015 was acquired by Ambarella (Santa Clara, CA). Tests will take place on specific streets in the Italian cities of Parma and Turin, with the assistance of a human supervisor. Meanwhile, the market for automotive radars is showing healthy dynamics: according to a report from market research firm Yole Développement, 24GHz radar held a large part of the 2018 market with US$2.2 billion and will grow slightly until 2020 for features like Blind Spot Monitor,  before being replaced by 79GHz high-resolution short-range radar that enables mapping of the entire car’s surroundings.

Powering hybrid and electric vehicles

Automotive was a key theme also at the recent PCIM show (Power Electronics, Intelligent Motion) held in Nuremberg, Germany, from May 7^th to May 9^th, as manufacturers of power devices are getting ready for the next generations of electric and hybrid vehicles. Infineon and Schweizer Electronic have developed a new technology for mild-hybrid cars (vehicle equipped with start-stop system and regenerative breaking): chip embedding for Power MOSFETs. It promises to improve the performance of 48V systems while reducing their complexity. Continental Powertrain will be the first player to adopt the technology. With chip embedding, the Power MOSFETs are no longer soldered onto a circuit board but integrated within. The resulting thermal benefits allow a higher power density, and board integration enables further improvements in system reliability. The benefits of 48V starter generators, compared to a 12V-based start-stop-system, include stopping the engine more often and for a longer duration, boosting acceleration, and recovering more energy when braking. Meanwhile, at the Electric & Hybrid Vehicle Technology Expo in Stuttgart, Germany, the French company Silicon Mobility announced its Olea App Inverter HE, a high efficiency inverter and electric motor control application optimized for its Olea FPCU (Field Programmable Control Unit) which claims to increase range of hybrid and electric vehicles up to 20% with the same battery capacity. The inverter integrates adaptive control algorithms which apply the most suitable control strategy upon requested power, motor angle, and speed, thus cutting energy losses in the power switches and in the electric motor.

PCB with integrated Power MOSFETs. Image credit: Infineon/Schweizer Electronic

Acquisitions, new fabs, and future TV sets

Other significant news from the last few days include Ansys acquiring DfR Solutions, the developer of Sherlock, an automated design reliability analysis software; Marvell acquiring Aquantia, a player in Multi-Gig Ethernet connectivity; Cree investing $1 billion to expand silicon carbide capacity at its U.S. headquarters in Durham, N.C.; and Huawei reportedly planning to introduce the world's first 5G 8K television (that is, a TV set receiving 8K video signal from the 5G network) as early as this year.

This entry was posted on Thursday, May 9th, 2019 at 3:11 pm. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.