Breaking the Bottleneck | Issue 28
[10/18/2023] Semi Scope 3, 3D Printing for Fusion, Deepmind RT-2, & New A24 Trailers
Breaking the Bottleneck is a weekly manufacturing technology newsletter with perspectives, interviews, news, funding announcements, and a startup database. For an older version of a discrete and continuous manufacturing market map click the link here!
Content I Enjoyed This Week 🏭🗞️🔬
News:
Decarbonizing Scope 3 Emissions in Semis [McKinsey]
Semiconductor companies play a critical role in the global technology ecosystem, producing microchips that power everything from smartphones to advanced automotive systems. Every step in the semiconductor manufacturing process, from wafer production to packaging, involves the use of fossil fuels and generates greenhouse gas emissions. To date, the semiconductor industry has made significant strides in reducing its environmental footprint by focusing on two primary categories of emissions: Scope 1 and Scope 2. However, these efforts only address 65% of emissions, the remaining 35% of emissions fall under Scope 3 emissions, which originate from the suppliers providing services, materials, and equipment necessary for chip manufacturing. The main sources of Scope 3 emissions in semiconductor fabs include Purchased Materials (62%), Maintenance Services and Spare Parts (22%), and Supplier Transportation (6%). However these emissions are hard to quantify because of the supplier & material distribution, the lack of transparency into associated materials, services, or suppliers (e.g. nitrogen trifluoride - NF3), and the industry's stringent requirements for materials (99.9% purity aluminum). Scope 3 emissions can be reduced with an increased focus by manufacturers on cooperative efforts with a select few suppliers, improved yield and waste reduction, material optimization/new substrates, and tradeoffs among product specifications.
Hydrogen as An Alternative Fuel [Industry Week]
Hydrogen is growing in interest across industrials offering several advantages, such as its abundance, zero emissions when used in fuel cells, and high specific energy, making it a potential game-changer for energy systems. However, there are still challenges prohibiting widespread adoption. First, green hydrogen production, achieved through electrolysis with renewable energy, requires a significant increase in electrolysis facilities and renewable energy sources to be sustainable. Secondly, handling and storing hydrogen safely are challenging due to its flammability and the need for proper infrastructure next to factories and logistics networks. Even with introducing the seven domestic hydrogen hubs new tools and engineering solutions are essential to driving adoption. Areas for opportunity include local energy storage, digital tools to design and optimize hydrogen-specific turbines/flow, and improved on-site and transmission infrastructure.
Tesla’s 4680 Battery Breakthrough [Electrek]
Tesla has manufactured its 20 millionth 4680 battery cell at Gigafactory Texas. The 4680 cell is considered crucial for Tesla's new vehicle programs, including the Cybertruck. Tesla unveiled this unique battery cell at Battery Day in 2020, claiming it could reduce costs by over 50%. However, the production ramp-up faced challenges. This announcement provides insight into the production rate, as Tesla previously kept this information confidential. Over the past 16 weeks, Tesla produced 10 million 4680 cells, averaging around 625,000 cells per week. The production rate may have increased since then, possibly reaching 800,000 cells per week. These 4680 cells are estimated to have a capacity of around 100 Wh, potentially producing about 80 MWh of battery cells per week. This quantity is sufficient to manufacture over 1,200 vehicles per week, considering a 65 kWh battery pack capacity.
Ford Taps IDRA for Gigapress [Teslarati]
Ford is collaborating with Italian company IDRA to explore the use of a Gigapress, a massive casting machine, for its electric vehicle (EV) manufacturing efforts. The Gigapress technology is familiar to Tesla enthusiasts, as Tesla has used IDRA's machinery to create large vehicle castings that improve manufacturing efficiency by reducing the number of parts needed. Ford is currently utilizing the Gigapress for research and development purposes and has not yet incorporated it into production units. The automaker aims to assess the suitability of the machine for its ongoing projects. IDRA showcased a Gigapress with Ford's logo at an industry event, while Hyundai is also considering using this technology, but there's no confirmation of their commitment to it. Tesla has been utilizing Gigapresses for several years, with 14 units in its possession, including two 9,000-ton units expected to be used for the production of the Cybertruck.
Using 3D Printed Metal/Ceramic Composites for Reactors [Voxel Matters]
China’s Factory Floor Is Moving—But Not to India or Mexico [WSJ]
Despite Western countries' efforts to reduce Chinese manufacturing dependency, China's global manufacturing grip remains strong. China's inland provinces are increasingly becoming major players in global manufacturing, challenging countries like Mexico, India, and Vietnam. This shift is driven by a search for cheaper labor, factory space, and tighter environmental regulations in China's coastal cities. As a result, exports from China's inland provinces have grown significantly, even outpacing the export growth of competing countries. China's central planners have long emphasized the importance of developing the interior to create an integrated national economy, and this trend continues today, with inland regions focusing on labor-intensive and low-value-added manufacturing while the coast concentrates on advanced industries.
Research:
RT-2: New Model Translates Vision and Language into Action [DeepMind]
A bit late but the Deepmind paper introduces Robotic Transformer 2 (RT-2), a novel vision-language-action (VLA) model designed to enhance robotic control. RT-2 combines knowledge from web-scale datasets and robotic data and builds upon the previous model, RT-1, which was trained on multi-task demonstrations using data collected from 13 robots over 17 months in an office kitchen environment. RT-2 exhibits improved generalization capabilities and a better understanding of semantics and visuals, even beyond the robotic data it was trained on. It can interpret new commands and perform rudimentary reasoning, such as categorizing objects or making high-level decisions. Additionally, it can perform multi-stage semantic reasoning and handle tasks like choosing an object for a specific purpose. To control robots, RT-2 leverages Vision-Language Models (VLMs) as its backbone, where actions are represented as tokens in the model's output. This representation enables the model to learn robotic actions from training data, aligning with language tokens. RT-2's ability to combine language and actions through chain-of-thought reasoning is also highlighted, demonstrating its capacity to plan long-horizon sequences of skills and predict robot actions.
Podcasts:
Manufacturing the New Artemis Lunar Drill
Deloitte on the Future of the Industrial Metaverse
Chart of the Week:
China’s Internal Trade is Growing faster than Exports from Rival Manufacturing Destinations
Track the Clean Manufacturing Factories Being Built Across the US
Manufacturing Deals
Formant - A cloud platform that helps robotics companies easily deploy, scale, and manage their fleets
$21 million [Venture Round] - Led by BMWiVentures and joined by Intel Capital, GS Futures SignalFire, Hillsven, Pelion Ventures, Holman, Ericsson, and Goodyear Ventures
Minds.ai - A company using deep learning to optimize semiconductor operations and planning
$5.3 million [Seed] - Led by MontaVista Capital and joined by Momenta Ventures
Azul 3D - A company that manufactures 3D printers to quickly print highly customizable structures
$15 million [Series A] - From DuPont, Beta Lab and GS Futures
Edgecortix - A company creating a high-speed yet very low-power, flexible edge AI-focused processor with a software-first approach
$20 million [Venture Round] - Led by Japan’s SBI Investment and Global Hands-On VC (GHOVC)
Phasio - a Singapore-based manufacturing interface that manages communication and workflow between manufacturers and customers
$2.5 million [Seed] - Led by Airtree Ventures and joined by 500 Global, Entrepreneur First, and Gattaca Ventures
Weekly Planned Downtime
New A24 Movie Trailers
Iron Claw
Priscilla