Tag: Semiconductor

  • The Asymmetric Indo-US Technology Agreement Points to India’s Weak R&D Culture

    The Asymmetric Indo-US Technology Agreement Points to India’s Weak R&D Culture

    Prime Minister Narendra Modi’s state visit to the USA resulted in four significant agreements and the visit is hailed as one of very important gains for India and Indo-US strategic partnership. The focus has been on defence industrial and technology partnership. Media and many strategic experts are seeing the agreements as major breakthroughs for technology transfers to India, reflecting a very superficial analysis and a lack of understanding of what really entails technology transfer. Professor Arun Kumar sees these agreements as a sign of India’s technological weakness and USA’s smart manoeuvring to leverage India for long-term defence and technology client. The visit has yielded major business gains for USA’s military industrial complex and the silicon valley. Post the euphoria of the visit, Arun Kumar says its time for India to carefully evaluate the relevant technology and strategic policy angles.

     

    The Indo-US joint statement issued a few days back says that the two governments will “facilitate greater technology sharing, co-development and co-production opportunities between the US and the Indian industry, government and academic institutions.” This has been hailed as the creation of a new technology bridge that will reshape relations between the two countries

    General Electric (GE) is offering to give 80% of the technology required for the F414 jet engine, which will be co-produced with Hindustan Aeronautics Limited (HAL). In 2012, when the negotiations had started, GE had offered India 58%. India needs this engine for the Light Combat Aircraft Mark 2 (LCA Mk2) jets.

    The Indian Air Force has been using LCA Mk1A but is not particularly happy with it. It asked for improvements in it. Kaveri, the indigenous engine for the LCA under development since 1986, has not been successful. The engine development has failed to reach the first flight.

    So, India has been using the F404 engine in the LCA Mk1, which is 40 years old. The F414 is also a 30-year-old vintage engine. GE is said to be offering 12 key technologies required in modern jet engine manufacturing which India has not been able to master over the last 40 years. The US has moved on to more powerful fighter jet engines with newer technologies, like the Pratt & Whitney F135 and GE XA100.

    India is being allowed into the US-led critical mineral club. It will acquire the highly rated MQ-9B high-altitude long-endurance unmanned aerial vehicles. Micron Technologies will set up a semiconductor assembly and test facility in Gujarat by 2024, where it is hoped that the chips will eventually be manufactured. The investment deal of $2.75 billion is sweetened with the Union government giving 50% and Gujarat contributing 20%. India is also being allowed into the US-led critical mineral club.

    There will be cooperation in space exploration and India will join the US-led Artemis Accords. ISRO and NASA will collaborate and an Indian astronaut will be sent to the International Space Station. INDUS-X will be launched for joint innovation in defence technologies. Global universities will be enabled to set up campuses in each other’s countries, whatever it may imply for atmanirbharta.

    What does it amount to?

    The list is impressive. But, is it not one-sided, with India getting technologies it has not been able to develop by itself.

    Though the latest technology is not being given by the US, what is offered is superior to what India currently has. So, it is not just optics. But the real test will be how much India’s technological capability will get upgraded.

    Discussing the New Economic Policies launched in 1991, the diplomat got riled at my complaining that the US was offering us potato chips and fizz drinks but not high technology, and shouted, “Technology is a house we have built and we will never let you enter it.”

    What is being offered is a far cry from what one senior US diplomat had told me at a dinner in 1992. Discussing the New Economic Policies launched in 1991, the diplomat got riled at my complaining that the US was offering us potato chips and fizz drinks but not high technology, and shouted, “Technology is a house we have built and we will never let you enter it.”

    Everyone present there was stunned, but that was the reality.

    The issue is, does making a product in India mean a transfer of technology to Indians? Will it enable India to develop the next level of technology?

    India has assembled and produced MiG-21 jets since the 1960s and Su-30MKI jets since the 1990s. But most critical parts of the Su-30 come from Russia. India set up the Mishra Dhatu Nigam in 1973 to produce the critical alloys needed and production started in 1982, but self-sufficiency in critical alloys has not been achieved.

    So, production using borrowed technology does not mean absorption and development of the technology. Technology development requires ‘know-how’ and ‘know-why’.

    When an item is produced, we can see how it is produced and then copy that. But we also need to know how it is being done and importantly, why something is being done in a certain way. Advanced technology owners don’t share this knowledge with others.

    Technology is a moving frontier

    There are three levels of technology at any given point in time – high, intermediate, and low.

    The high technology of yesterday becomes the intermediate technology of today and the low technology of tomorrow. So, if India now produces what the advanced countries produced in the 1950s, it produces the low-technology products of today (say, coal and bicycles).

    If India produces what was produced in the advanced countries in the 1980s (say, cars and colour TV), it produces the intermediate technology products of today. It is not to say that some high technology is not used in low and intermediate-technology production.

    The high technologies of today are aerospace, nanotechnology, AI, microchips and so on. India is lagging behind in these technologies, like in producing passenger aircraft, sending people into space, making microchips, quantum computing, and so on.

    The advanced countries do not part with these technologies. The World Trade Organisation, with its provisions for TRIPS and TRIMS (Trade-Related Aspects of Intellectual Property Rights and Trade-Related Investment Measures), consolidated the hold of advanced countries on intermediate and low technologies that can be acquired by paying royalties. But high technology is closely held and not shared.

    Advancements in technology

    So, how can nations that lag behind in technology catch up with advanced nations? The Nobel laureate Kenneth Arrow pointed to ‘learning by doing’ – the idea that in the process of production, one learns.

    So, the use of a product does not automatically lead to the capacity to produce it, unless the technology is absorbed and developed. That requires R&D.

    Schumpeter suggested that technology moves through stages of invention, innovation and adaptation. So, the use of a product does not automatically lead to the capacity to produce it, unless the technology is absorbed and developed. That requires R&D.

    Flying the latest Airbus A321neo does not mean we can produce it. Hundreds of MiG-21 and Su-30 have been produced in India. But we have not been able to produce fighter jet engines, and India’s Kaveri engine is not yet successful. We routinely use laptops and mobile phones, and they are also assembled in India, but it does not mean that we can produce microchips or hard disks.

    Enormous resources are required to do R&D for advanced technologies and to produce them at an industrial scale. It requires a whole environment which is often missing in developing countries and certainly in India.

    Enormous resources are required to do R&D for advanced technologies and to produce them at an industrial scale. It requires a whole environment which is often missing in developing countries and certainly in India.

    Production at an experimental level can take place. In 1973, I produced epitaxial thin films for my graduate work. But producing them at an industrial scale is a different ballgame. Experts have been brought from the US, but that has not helped since high technology is now largely a collective endeavour.

    For more complex technologies, say, aerospace or complex software, there is ‘learning by using’. When an aircraft crashes or malware infects software, it is the producer who learns from the failure, not the user. Again, the R&D environment is important.

    In brief, using a product does not mean we can produce it. Further, producing some items does not mean that we can develop them further. Both require R&D capabilities, which thrive in a culture of research. That is why developing countries suffer from the ‘disadvantage of a late start’.

    A need for a focus on research and development

    R&D culture thrives when innovation is encouraged. Government policies are crucial since they determine whether the free flow of ideas is enabled or not. Also of crucial importance is whether thought leaders or sycophants are appointed to lead institutions, whether criticism is welcomed or suppressed, and whether the government changes its policies often under pressure from vested interests.

    Unstable policies increase the risk of doing research, thereby undermining it and dissuading the industry. The result is the repeated import of technology.

    The software policy of 1987, by opening the sector up to international firms, undermined whatever little research was being carried out then and turned most companies in the field into importers of foreign products, and later into manpower suppliers. Some of these companies became highly profitable, but have they produced any world-class software that is used in daily life?

    Expenditure on R&D is an indication of the priority accorded to it. India spends a lowly 0.75% of its GDP on R&D. Neither the government nor the private sector prioritises it. Businesses find it easier to manipulate policies using cronyism. Those who are close to the rulers do not need to innovate, while others know that they will lose out. So, neither focus on R&D.

    Innovation also depends on the availability of associated technologies – it creates an environment. An example is Silicon Valley, which has been at the forefront of innovation. It has also happened around universities where a lot of research capabilities have developed and synergy between business and academia becomes possible.

    This requires both parties to be attuned to research. In India, around some of the best-known universities like Delhi University, Allahabad University and Jawaharlal Nehru University, coaching institutions have mushroomed and not innovative businesses. None of these institutions are producing any great research, nor do businesses require research if they can import technology.

    A feudal setup

    Technology is an idea. In India, most authority figures don’t like being questioned. For instance, bright students asking questions are seen as troublemakers in most schools. The emphasis is largely on completing coursework for examinations. Learning is by rote, with most students unable to absorb the material taught.

    So, most examinations have standard questions requiring reproduction of what is taught in the class, rather than application of what is learned. My students at JNU pleaded against open-book exams. Our class of physics in 1967 had toppers from various higher secondary boards. We chose physics over IIT. We rebelled against such teaching and initiated reform, but ultimately most of us left physics – a huge loss to the subject.

    Advances in knowledge require critiquing its existing state – that is, by challenging the orthodoxy and status quo. So, the creative independent thinkers who generate socially relevant knowledge also challenge the authorities at their institutions and get characterised as troublemakers. The authorities largely curb autonomy within the institution and that curtails innovativeness.

    In brief, dissent – which is the essence of knowledge generation – is treated as a malaise to be eliminated. These are the manifestations of a feudal and hierarchical society which limits the advancement of ideas. Another crucial aspect of generating ideas is learning to accept failure. The Michelson–Morley experiment was successful in proving that there is no aether only after hundreds of failed experiments.

    Conclusion

    The willingness of the US to provide India with some technology without expecting reciprocity is gratifying. Such magnanimity has not been shown earlier and it is obviously for political (strategic) reasons. The asymmetry underlines our inability to develop technology on our own. The US is not giving India cutting-edge technologies that could make us a Vishwaguru.

    India needs to address its weakness in R&D. As in the past, co-producing a jet engine, flying drones or packaging and testing chips will not get us to the next level of technology, and we will remain dependent on imports later on.

    This can be corrected only through a fundamental change in our R&D culture that would enable technology absorption and development. That would require granting autonomy to academia and getting out of the feudal mindset that presently undermines scientific temper and hobbles our system of education.

     

    This article was published earlier in thewire.in

    Feature Image Credit: thestatesman.com

     

  • The Geopolitical Consolidation of Artificial Intelligence

    The Geopolitical Consolidation of Artificial Intelligence

    Key Points

    • IT hardware and Semiconductor manufacturing has become strategically important and critical geopolitical tools of dominant powers. Ukraine war related sanctions and Wassenaar Arrangement regulations invoked to ban Russia from importing or acquiring electronic components over 25 Mhz.
    • Semi conductors present a key choke point to constrain or catalyse the development of AI-specific computing machinery.
    • Taiwan, USA, South Korea, and Netherlands dominate the global semiconductor manufacturing and supply chain. Taiwan dominates the global market and had 60% of the global share in 2021. Taiwan’s one single company – TSMC (Taiwan Semiconductor Manufacturing Co), the world’s largest foundry, alone accounted for 54% of total global revenue.
    • China controls two-thirds of all silicon production in the world.
    • Monopolisation of semiconductor supply by a singular geopolitical bloc poses critical challenges for the future of Artificial Intelligence (AI), exacerbating the strategic and innovation bottlenecks for developing countries like India.
    • Developing a competitive advantage over existing leaders would require not just technical breakthroughs but also some radical policy choices and long-term persistence.
    • India should double down over research programs on non-silicon based computing with a national urgency instead of pursuing a catch-up strategy.

    Russia was recently restricted, under category 3 to category 9 of the Wassenaar Arrangement, from purchasing any electronic components over 25MHz from Taiwanese companies. That covers pretty much all modern electronics. Yet, the tangibles of these sanctions must not deceive us into overlooking the wider impact that hardware access and its control have on AI policies and software-based workflows the world over. As Artificial Intelligence technologies reach a more advanced stage, the capacity to fabricate high-performance computing resources i.e. semiconductor production becomes key strategic leverage in international affairs.

    Semiconductors present a key chokepoint to constrain or catalyse the development of AI-specific computing machinery. In fact, most of the supply of semiconductors relies on a single country – Taiwan. The Taiwan Semiconductor Manufacturing Corporation (TSMC) manufactures Google’s Tensor Processing Unit (TPU), Cerebras’s Wafer Scale Engine (WSE), as well as Nvidia’s A100 processor. The following table provides a more detailed1 assessment:

    Hardware Type

    AI Accelerator/Product Name

    Manufacturing Country

    Application-Specific Integrated Circuits (ASICs)

    Huawei Ascend 910

    Taiwan

    Cerebras WSE

    Taiwan

    Google TPUs

    Taiwan

    Intel Habana

    Taiwan

    Tesla FSD

    USA

    Qualcomm Cloud AI 100

    Taiwan

    IBM TrueNorth

    South Korea

    AWS Inferentia

    Taiwan

    AWS Trainium

    Taiwan

    Apple A14 Bionic

    Taiwan

    Graphic Processing Units (GPUs)

    AMD Radeon

    Taiwan

    Nvidia A100

    Taiwan

    Field-Programmable Gate Arrays (FPGAs)

    Intel Agilex

    USA

    Xilinx Virtex

    Taiwan

    Xilinx Alveo

    Taiwan

    AWS EC2 FI

    Taiwan

    As can be seen above, the cake of computing hardware is largely divided in such a way that the largest pie holders also happen to form a singular geopolitical bloc vis-a-vis China. This further shapes the evolution of territorial contests in the South China Sea. This monopolisation of semiconductor supply by a singular geopolitical bloc poses critical challenges for the future of Artificial Intelligence, especially exacerbating the strategic and innovation bottlenecks for developing countries like India. Since the invention of the transistor in 1947, and her independence, India has found herself in an unenviable position where there stands zero commercial semiconductor manufacturing capacity after all these years while her office-bearers continually promise of leading in the fourth industrial revolution.

    Bottlenecking Global AI Research

    There are two aspects of developing these AI accelerators – designing the specifications and their fabrication. AI research firms first design chips which optimise hardware performance to execute specific machine learning calculations. Then, semiconductor firms, operating in a range of specialities and specific aspects of fabrication, make those chips and increase the performance of computing hardware by adding more and more transistors to pieces of silicon. This combination of specific design choices and advanced hardware fabrication capability forms the bedrock that will decide the future of AI, not the amount of data a population is generating and localising.

    However, owing to the very high fixed costs of semiconductor manufacturing, AI research has to be focused on data and algorithms. Therefore, innovations in AI’s algorithmic efficiency and model scaling have to compensate for a lack of equivalent situations in the AI’s hardware. The aggressive consolidation and costs of hardware fabrication mean that firms in AI research are forced to outsource their hardware fabrication requirements. In fact, as per DARPA2, because of the high costs of getting their designs fabricated, AI hardware startups do not even receive much private capital and merely 3% of all venture funding between 2017-21 in AI/ML has gone to startups working on AI hardware.

    But TSMC’s resources are limited and not everyone can afford them. To get TSMC’s services, companies globally have to compete with the likes of Google and Nvidia, therefore prices go further high because of the demand side competition. Consequently, only the best and the biggest work with TSMC, and the rest have to settle for its competitors. This has allowed this single company to turn into a gatekeeper in AI hardware R&D. And as the recent sanctions over Russia demonstrate, it is now effectively playing the pawn which has turned the wazir in a tense geopolitical endgame.

    Taiwan’s AI policy also reflects this dominance in ICT and semiconductors – aiming to develop “world-leading AI-on-Device solutions that create a niche market and… (make Taiwan) an important partner in the value chain of global intelligent systems”.3 The foundation of strong control over the supply of AI hardware and also being #1 in the Global Open Data Index, not just gives Taiwan negotiating leverage in geopolitical competition, but also allows it to focus on hardware and software collaboration based on seminal AI policy unlike most countries where the AI policy and discourse revolve around managing the adoption and effects of AI, and not around shaping the trajectory of its engineering and conceptual development like the countries with hardware advantage.

    Now to be fair, R&D is a time-consuming, long-term activity which has a high chance of failure. Thus, research focus naturally shifts towards low-hanging fruits, projects that can be achieved in the short-term before the commissioning bureaucrats are rotated. That’s why we cannot have a nationalised AGI research group, as nobody will be interested in a 15-20 year-long enterprise when you have promotions and election cycles to worry about. This applies to all high-end bleeding-edge technology research funding everywhere – so, quantum communications will be prioritised over quantum computing, building larger and larger datasets over more intelligent algorithms, and silicon-based electronics over researching newer computing substrates and storage – because those things are more friendly to short-term outcome pressures and bureaucracies aren’t exactly known to be a risk-taking institution.

    Options for India

    While China controls 2/3 of all the silicon production in the world and wants to control the whole of Taiwan too (and TSMC along with its 54% share in logic foundries), the wider semiconductor supply chain is a little spreadout too for any one actor’s comfort. The leaders mostly control a specialised niche of the supply chain, for example, the US maintains a total monopoly on Electronic Design Automation (EDA) software solutions, the Netherlands has monopolised Extreme UltraViolet and Argon Flouride scanners, and Japan has been dishing out 300 mm wafers used to manufacture more than 99 percent of the chips today.4 The end-to-end delivery of one chip could have it crossing international borders over 70 times.5 Since this is a matured ecosystem, developing a competitive advantage over existing leaders would require not just proprietary technical breakthroughs but also some radical policy choices and long term persistence.

    It is also needless to say that the leaders are also able to attract and retain the highest quality talent from across the world. On the other hand, we have a situation where regional politicians continue cribbing about incoming talent even from other Indian states. This is therefore the first task for India, to become a technology powerhouse, she has to, at a bare minimum, be able to retain all her top talent and attract more. Perhaps, for companies in certain sectors or of certain size, India must make it mandatory to spend at least X per cent of revenue on R&D and offer incentives to increase this share – it’ll revamp things from recruitment and retention to business processes and industry-academia collaboration – and in the long-run prove to be a lot more socioeconomically useful instrument than the CSR regulation.

    It should also not escape anyone that the human civilisation, with all its genius and promises of man-machine symbiosis, has managed to put all its eggs in a single basket that is also under the constant threat of Chinese invasion. It is thus in the interest of the entire computing industry to build geographical resiliency, diversity and redundancy in the present-day semiconductor manufacturing capacity. We don’t yet have the navy we need, but perhaps in a diplomatic-naval recognition of Taiwan’s independence from China, the Quad could manage to persuade arrangements for an uninterrupted semiconductor supply in case of an invasion.

    Since R&D in AI hardware is essential for future breakthroughs in machine intelligence – but its production happens to be extremely concentrated, mostly by just one small island country, it behoves countries like India to look for ways to undercut the existing paradigm of developing computing hardware (i.e. pivot R&D towards DNA Computing etc) instead of only trying to pursue a catch-up strategy. The current developments are unlikely to solve India’s blues in integrated circuits anytime soon. India could parallelly, and I’d emphatically recommend that she should, take a step back from all the madness and double down on research programs on non-silicon-based computing with a national urgency. A hybrid approach toward computing machinery could also resolve some of the bottlenecks that AI research is facing due to dependencies and limitations of present-day hardware.

    As our neighbouring adversary Mr Xi says, core technologies cannot be acquired by asking, buying, or begging. In the same spirit, even if it might ruffle some feathers, a very discerning reexamination of the present intellectual property regime could also be very useful for the development of such foundational technologies and related infrastructure in India as well as for carving out an Indian niche for future technology leadership.

    References:

    1. The Other AI Hardware Problem: What TSMC means for AI Compute. Available at https://semiliterate.substack.com/p/the-other-ai-hardware-problem

    2. Leef, S. (2019). Automatic Implementation of Secure Silicon. In ACM Great Lakes Symposium on VLSI (Vol. 3)

    3. AI Taiwan. Available at https://ai.taiwan.gov.tw/

    4. Khan et al. (2021). The Semiconductor Supply Chain: Assessing National Competitiveness. Center for Security and Emerging Technology.
    5. Alam et al. (2020). Globality and Complexity of the Semiconductor Ecosystem. Accenture.