Deeptech Round Up for October 5th

Deep tech signals that indicate the future may be different from today:

The epitome of deep tech, the flying car, is slowing becoming a reality with Alef Aeronautics impressing the Detroit Auto show with its futuristic Model A. The flying car is designed for use by the general public and could be in the skies as early as 2025. But with a $300,000 price tag and regulatory hurdles to overcome it is unlikely we will see widespread adoption soon.

Meanwhile Meta are still pursuing their metaverse dreams with a recent interview between Mark Zuckerberg and Lex Fridman showing how quickly their software has improved. Not only do Meta avatars now have legs but they can deliver near photo-realism levels of detail. Whilst Mark didn’t wear one, shortly after the interview, Meta announced the launch of their Quest 3, a mixed reality headset clearly aimed at the mass market with a price tag of $499.99.

A mixed reality world opens up all sorts of exciting possibilities and use cases but also has big implications for the human condition. Recent research from Cornell and Brown university found that in interactions between AR wearers and non-wearers, the non-wearers were at a significant disadvantage.

ESAs next Vega flight will carry 3 ANSER CubeSats that are designed to orbit in formation acting like a much larger single satellite to image the waters around Iberia.

University of Washington researchers have developed battery-free robots that change shape in mid-air to stabilise and lengthen their descent. The micro-gliders weight just 400g and are designed to carry tiny sensors to measure temperature, humidity and other environmental conditions.

Mistral AI has made its first large language model (LLM) free for everyone. Although not quite as capable as GPT-4, the model has delivered impressive benchmarks given its relatively small size. It is also remarkable how quickly Mistral were able to launch the model given the company did not exist 7 months ago.

Despite massive progress with LLMs in recent months they still suffer from Hallucinations. DARPA has now selected 4 teams to build trustworthy AI systems. The Assured Neuro Symbolic Learning and Reasoning (ANSR) program focuses on combining machine learning and traditional symbolic reasoning to build robust and reliable autonomous systems that can hopefully not succumb to hallucination.

Deeptech Reads

Longer form content to dive in deep with:

Scientists at the Princeton Plasma Physics Laboratory have made a breakthrough that has huge potential to improve the efficiency of tokamak fusion devices by mitigating the impact of runaway electrons that can significantly damage the walls of a Tokamak device.

There were many fascinating discussions on deep tech at TechCrunch Disrupt 2023. The following is an interesting one on the potential and limitations of applying quantum technologies to espionage:

The more I learn, the more excited I become about the potential of quantum computing but there are many challenges. This article does is great for comparing the current challenges in quantum computing with the history of the digital computer.

The fundamental building block of a quantum computer is the qubit. But not all qubits are the same. In fact, today there are three main qubit technologies each with their own pros and cons and it is not yet obvious which is the most likely technology to achieve commercial scale.

Early digital computers were plagued by errors but in hindsight these were fairly trivial to overcome through digitisation, precision parts, and error correction codes. Error correction in quantum computing is a much more intractable problem. But progress is being made. MIT researchers using a type of superconducting qubit called a fluxonium, which has much longer lifespans than a conventional qubit, have achieved 99.9% accuracy with two-qubit gates and 99.99% accuracy with a single qubit gate. This is very impressive but shades in comparison with a modern classical computer which can achieve an equivalent accuracy of about 10-18.

However, just because real quantum computers are difficult to build does not mean we can’t benefit from advances in quantum computing. In fact, any Turing Complete classical computer can, in principle, replicate any quantum calculation. It is just that the classical computer will be much slower at doing so. However, the race for AI is driving huge developments in GPU capabilities. This is enabling companies like Sandbox AQ to run useful quantum equations and software on GPUs rather than on real quantum computing hardware.

According to Roland Berger, the per capita computational demand in major countries is set to increase 20-fold reaching an impressive 10 TFLOPs by 2035. To satiate this level of computational power and balance the associated cost and usability requirements will likely require a shift to distributed computing.

The latest buzzword in distributed computing is “Fog Computing”. The idea is to create a system of cloud data centres, edge devices and intermediary “fog nodes” in a distributed network that can optimise efficiency and reduce latency. Personally, whilst I think the concept is great I am not a fan of the name and the actual implementation of such a system seems quite challenging to me.

One element that could make Fog Computing more viable is Conflict-Free Replicated Data (CRDTs). These are a class of data structure designed to ensure that the data stays consistent across all replicas in a distributed system. Data can be modified even in the presence of a network partition or a node being offline and the CRDTs will ensure data consistency once the node is reconnected.

Whilst there is much hype, green hydrogen won’t happen without adequate levels of investment and support as a recent report by the IEA makes clear.

As made clear by the EUs recent announcement around, critical technology, deep tech is increasingly becoming a core part of international relations:

And as the US doubles down on an industrial renaissance LA is emerging as a hub of hardtech innovation

Deeptech Deals

A round up of deep tech deals and new programs:

  • Oorja has raised a $1.5M pre-series A round led by Micelio Fund with participation from Capital-AJava CapitalAnicut Capital and Angels. Oorja was launched in 2022 in Bengaluru and uses predictive model to design battery solutions for EVs.
  • Applied Ventures, the VC arm of Applied Materials has selected 7 startups for the 4th cohort of its Applied Startup Technology & Research Accelerator (ASTRA). The selected startups include:
  • PierSight has raised $600,000 in pre-seed funding led by All In Capital. The graduates from the Techstars Space Accelerator program are combining Synthetic Aperture Radar and Automatic Identification Systems to provide persistent surveillance of the marine environment to combat illegal fishing, environmental damage and smuggling.
  • The UK Space Agency has launched up to £65M in funding as part of the UKs National Space Innovation Program. The first call is out and applications close at midday on 17th November.
  • Intuitive Machines has moved into a new $40M HQ at the Houston Spaceport to support NASAs $93Bn Artemis program to return astronauts to the moon by 2024 and eventually send humans to Mars.
  • Nucor Corporation has announced a collaboration agreement with Helion to develop a 500MW fusion plant to power Nucor’s steelmaking facility. The agreement includes a direct investment of $35M into Helion and a target of achieving operations by 2030.
  • Harbringer Health has raised a $140M series B round to help it complete a 10,000 subject clinical trial of its blood based cancer screening test. If successful Harbringer aims to launch a laboratory test in 2025.
  • Lyten has raised $200M to scale up its graphene battery technology. The use of graphene enables lithium-sulphur batteries that do not use nickel, cobalt or manganese resulting in a projected 50% lower cost of materials. The company aims to start shipping batteries in early 2024.
  • The Rice Alliance Energy Tech Venture Forum has named its 10 “most promising” energy startups:
    • 1s1 Energy – next generation electrolyzers for low-cost green hydrogen
    • Ayrton Energy – Liquid hydrogen storage technology without the need for cryogenics and pressure
    • Carbonloop – Converting CO2 into liquid fuel and other high-value products
    • Mantel – low-cost, thermally efficient, carbon capture technology
    • Mars Materials – CO2 sequestration into feedstock for carbon fibre and wastewater treatment markets
    • Mirico – Measuring real time emissions from operations
    • Mobilus Labs – empowering frontline workers with a digital communication platform
    • Numat – reducing environmental damage through precision chemistry
    • Polystyvert – Polystyrene recycling technology
    • Protein Evolution – breaking down polyester textile waste through green chemistry
  • Honeywell has invested in $27.5M into iron flow battery company ESS Tech to help in the development of long term energy storage.