(The Report outlined below is based on a talk given by Prof Manan Suri, Department of Electrical Engineering, IIT, Delhi at the VIF on 23 November 2023. It is enriched by the valuable inputs provided by the senior members from the armed services, distinguished scientists, industry representatives, academia and strategic experts in addition to senior expert group members from the VIF. Report has been prepared by Dr. Saroj Bishoyi, VIF)

The world is witnessing an inflection point in the realm of semiconductor technology and its applications. The factors that contribute to this phenomenon are the saturation of Moore’s law, and the ease in which we generate enormous amounts of data. The nature of existing data intensive applications is such that, excellence in computational performance cannot be achieved based on raw transistor scaling or increasing the number of processing cores alone. Hence, a fundamental shift in the vastly successful Von Neumann computational paradigm is needed to overcome the bottlenecks associated with data-intensive real-time applications. In this endeavour, a research group at IIT, Delhi led by Prof Manan Suri, is working on the next generation of semiconductor memory technology. They are working on exploiting the characteristics of emerging Non Von Neumann (NVM) semiconductor technology for a multitude of novel applications such as data storage, low-power bio-inspired computing and cyber-security.

Semiconductor Technology

Semiconductor technology is a critical domain for security of the country. Generally, emphasis in semiconductor domain has been on faster and more efficient processing or computation. Storage technology has not received enough attention, even though it is of equal importance for fidelity of data, input-output speeds, secured operation etc. Therefore, Prof Suri’s research group at IIT, Delhi decided to focus on semiconductor storage technology, i.e. memory devices. All the efforts are being directed towards the creation of niche memory technology rather than computing or processing. Under IIT’s incubation and innovation programme, there are various programmes and policies for faculty members to take technologies out of the lab for the domestic users. Under this programme, Prof Suri and his team incubated an IIT Delhi startup, i.e., CYRAN, in 2018. Since then they have been working with various users who require cutting edge deep-technology solutions.

Deep-Tech Self-Reliance

The startup CYRAN, which stands for Cyber Raksha AI Nano-electronics, has contributed the deep-tech solutions towards Atmanirbhar Bharat or Self-Reliant India initiates. It has provided cyber solution both in terms of physical and software side. On the cyber-physical security hardware solution, the group has provided solutions in terms of HW crypto primitives, storage security and HW forensics. On the software side, it has provided AI HW-SW edge solutions such as AI authentication/Biometrics, AI Geospatial, AI industry 4.0 and AI EdTech.

They deployed first semiconductor hardware (HW) security solution with the government of India in 2017. They deployed first indigenous artificial intelligence (AI) solution with the government of India in 2018. On 7th October 2023, the AI technology developed by CYRAN for Innovation for Defence Excellence (iDEX) was added in the fifth positive indigenisation list by the MoD. The team received several recognitions recently, including National Technology Startup Award in 2021, Dare to Dream 2.0 award 2021 from DRDO, and Defence Minister’s award for Excellence in Defence and Aerospace Sector in 2022.

Managing Data Problem

The data problem is multi-dimensional and many factors contribute to it. There is a lot of data or information that needs to be managed in a meaningful way. For finding long-term solution, there is necessity to go into the deeper aspects of the data problem. On managing the data, there are four essential verticals such as i) Quest for a Digital Empire, ii) the Nature of Our World, the real world where we live in; iii) Quest for Intelligent Machines, and iv) the Nature of Technology Itself. These are closely linked to how the semiconductor technology evolves.

Quest for a Digital Empire

So far as the Quest for a Digital Empire is concerned, everything that we look at such as internet of things (IoT), e-Governance, healthcare, defence, security, space, smart cities, social media, retail and finance, industry 4.0, etc. we are in a quest of creating digital empire in every field of our life. This leads to phenomenal amount of data creation. There are industry estimates which are fairly accurate. A few years ago, it was projected that the total amount of data to be generated by 2021 would be 62 thousand billion gigabytes. Obviously this new data has been generated and the New Data Economy is accelerating data creation. So when it comes to storing data reliably, the semiconductor industry needs to produce higher capacity, compact and energy efficient data storage solutions. This also includes devices like Pen drive, SSD (Solid State Drive), HDD (Hard Disk Drive), etc.

The Nature of Our World Driven by Data

The nature of our world, everything including security, health, climate change, basic sciences, and weather modelling, etc. are driven by data. The nature of the problems are such that unless we generate a lot of data, they cannot be dealt with. If we start with small volume of data, we cannot reach to a meaningful solution. The scale of magnitude can be assessed by several published examples. It is published information that defence drones generate Petabytes of data per day. Surveillance cameras globally generate exabytes of data per day. So security concerns alone generate massive amount of data that is required to be stored and processed. Many other types of data generation can be added to the list. With improvements in sensor, data communication and many other aspects of requirements and technologies, the amount of data generated per day is set to grow. This makes data management and processing an institutional, rather than an individual, requirement. So the right kind of focus is required on developing semiconductor technology for storage devices also.

Quest for Intelligent Machines

The first stage is to store huge data in real physical world. Once the data is stored, the next stage is to process it or to generate meaningful information from this data, leading the Quest for Intelligent Machines. The expectation of any common user is to interact with more and more intelligent machines in their ecosystem in their daily lives whether we call it AI or GPT (Generative Pre-trained Transformer). The expectation is that the machines should have more and more capability to process data and faster. Consequently, these machines have become more autonomous and project back a lot of raw information or data. For instance, a self-driving car generates 40 terabyte (TB) data with just 8 hours of drive. So if we calculate the total number of self-driving cars used in a specific area or State, how many hours these are driven and multiply with how much data is generated that would be humongous. Interestingly, approximately 90% of energy is lost on data management, and only about 10 % on computation. This sets a clear direction for researchers to focus on efficient memories in quest for intelligent machines. Hence, the focus is now moving more on storage or memory paradigm.

The Nature of Technology Itself

This is a positive feedback loop. According to the Moore’s law, every year more and more semiconductors can be manufactured in the same amount of the silicon and this leads to an economy of scale. So sensors, processors, etc. all these things become cheaper and easier. That itself led to generation of more data and information. It is like feeding into itself. So today the cheapest thing to do is to generate data by not doing anything, walking around, by sitting somewhere, one ends up generating data for free. Even this can be monetised with one’s knowledge. One holding a phone and swinging around already generated data. So it is easiest thing to do.

Memory Centric Neuromorphic Systems

The concept of neuromorphic computing or memory based computing can be understood from the semiconductor based computing systems. As an example, the pen drive can store data (images, files, etc.) but cannot extract information from it unless it is connected to a computer. Unlike this, the human brain is able to store and process data simultaneously. Human brain is a massive parallel storage and processing unit with 1011 Neurons and 1015 Synapses, all in about 2 L Volume and needing low power of about 20 W. Computational neuroscience is the study of the learning mechanisms, rules and information processing capabilities and techniques in real organisms and then mimicking them into AI engines, also called artificial machines. Neuromorphic computing is the latest and state of art in AI. The research group of Prof Suri at IIT Delhi is one of the leading groups globally working on neuromorphic systems.

Power Consumption Landscape

When one looks at the power consumption landscape in the context of conventional hardware and neuromorphic hardware, the large dams are at a scale of gigawatts. The present day supercomputer chip cluster, not analogous to neural function, which needs its own power centre to power its own data centre, uses several megawatts, whereas the Wind Turbines rage in around few megawatts scale. Coming down the order, the state of the art CPU rating at few 100s of watts. Coming down by an order further, the biological computing is 20 watts. Furthermore, laser in CD/DVD player uses 5-10 milliwatts, IBM’s True North uses tens of milliwatts (1 million artificial neurons, or 256 million synapses). One old device made by humans, which is very low in spectrum that is hearing aid, consumes less than 1 milliwatt.

The objective is to take both the storage and computing process to that sub-milliwatt stage. The idea is to come down to the natural level of energy dissipation because the tech industries want to have more powerful computational power. In this quest, the industries emphasise on the sustainability of the technology. Moreover, the edge intelligence system like drone systems, autonomous underwater platforms, aerial platforms, space platforms, satellites, nano-satellites, etc. with storage system in them would run with low power, which will not create battery issues, thereby will increase the lifetime issues.

Non Von Neumann (NVN) Computing

The leading tech industries around the world such as IBM, Intel, Google, NVidia and Cerebras are investing a lot in this new computing architecture and new computing chips. Their effort is to go in the direction of new form of computing that is Non Von Neumann (NVN) Computing or the bio inspired computing. This is definitely setting the tone in the areas of investing in the new hardware architecture, and new fundamental way of computing that includes the storage or memory aspects. Importantly, Amazon, Google and Microsoft were never into the hardware manufacturing, especially never into the chip design. But they are now giving good competition to the traditional chip design tech companies by having their own chip design team, and computing architecture team. They are now making efforts to have their in-house designed computing chips that consume low power and are compliant with state of art security aspects.

Memory Technology

The memory technology today appears as a science fiction, given that 1TB SSD can be stored in a size of stamp. But if one looks into the microscopic level, it is actually a three dimensional chip or three dimensional structure. Memory technology has now reached a matured stage. In addition, the semiconductor materials are important for having a meaningful stake in semiconductor technology. It is known that there are few countries who control the supply chain of the semiconductor materials and geopolitics also controls the supply chain. India does not have much control over semiconductor materials or manufacturing capability. So access to semiconductor technology and materials is critically important to set up semiconductor fabs in India.

Economic Security

The worldwide semiconductor imports have crossed the oil and related imports. Semiconductors are the building blocks of electronic devices and use of electronics such as tablets, cell-phones, set-top boxes, digital TVs, automotive, medical, game consoles, military, etc. have significantly increased in the last few years. For joint development and production of semiconductor technology, unless there is compelling economic factor or incentive, no tech industry or country will agree to start this venture. So joint production is purely economics. How much we can offer as a consumer that also will play into this.

Geopolitical Aspects

The semiconductor foundries are currently located in select countries and the global semiconductor supply chain is dominated by them. At present, the relevant players are US, Japan, Taiwan, South Korea, China and some sites in Europe. Unfortunately, India so far has only one Semi-Conductor Laboratory (SCL), Mohali. However, the global semiconductor supply chain is very complex issue. The typical semiconductor production process spans multiple countries and goes through different stages of production processes: silicon ingots cut into wafers; bare wafer into fab wafer; fab wafer sorted, cut into die; die are assembled packaged, tested; final product shipped for inventory; chip integrated into consumer goods by end production manufacturer; and then customer buys end product.

It is important to note that 70 % of semiconductors pass through Taiwan/China during production process, which raises concern in Washington, when it factors China’s policy toward Taiwan. Thus, when we talk about zero trust in critical electronics, security will not be ensured only through software but hardware part of it is also important. So having some minimum end-to-end semiconductor capability locally or by authorised companies for at least strategic or security applications is absolutely essential which is not monitored through this global supply chains. So India needs to break into the global supply chain by developing its own semiconductors.

Hardware Layer Threats to Zero Trust Model

The hardware layer threats to zero trust model can come in several ways: Trojan, backdoors and untrusted foundry, counterfeit ICs, physical attack, side-channel, fault injection, reverse engineering, and fake parts. The research group at the IIT, Delhi also studies hardware security. They look into some specific areas such as side-channel, fault injection, among others. This is their primary motivation in this field: nature of data, nature of computing, which they are trying to solve these problems. The factors like geo-economics, geopolitics, and security of supply chain should also play in any policy framework.

Case Studies

The research group at the IIT, Delhi extensively works on the advanced semiconductor memory technology. The kind of technology that can store more and more data in the time to come. This is very interdisciplinary cycle because it starts with material, from material it goes into electrical engineering, then it goes into some modelling, circuit design, and then it goes into applications. For a meaningful outcome, whole cycle needs to be looked into. If any policy framework looks into only one or two layers, then it will only achieve short-term outcome. Other areas that the research group works is on Neuromorphic Computing and Sensing. For this area of work, they got Dare to Dream award from the DRDO in 2021. A sensor with neuromorphic computing helps saving energy and storage by removing redundant parts. They are also working on low power bio-inspired AI – Neuromorphic chips; Low power computing; commercial applications such as Low power edge – AI chips; hardware security; and, neuromorphic applications for healthcare. So staying relevant for the next 15 years from the technical side, there is a need of developing future relevant technologies like 3D chip, new materials, and new architecture. Every chip that is available today is a 2D chip. These will become 3D chips in future with non-silicon materials.

India Context

At present, the bodies like Semi-Conductor Laboratory (SCL), Mohali; Sahasra Semiconductors Pvt. Ltd (SSPL), Bhiwadi; Society for Integrated Circuit Technology and Applied Research (SITAR), Bengaluru; the Indian Institutes of Technology (IITs); the Indian Institute of Science (IISc), Bangalore have some capability and capacity in the semiconductor sector. There is, however, need to upscale them. They need to come at par with 3D chips that we are currently aiming at. Then there are programmes like India Semiconductor Mission (ISM), which is helping to build a semiconductor ecosystem in the country and programmes like the Chips to Startup (C2S). The Ministry of Electronics and Information (MeitY) has sought applications from academia, R&D organisations, start-ups and MSMEs under its C2S Programme. The government has also launched the production-linked incentive (PLI) scheme to encourage semiconductor manufacturing within India. While there is strong emphasis on training and skilling; and the country is providing abundant semiconductor design services without having product/OEM players in India or of Indian origin. In the long run if we just rely on services for the semiconductor sector as a country we may not gain much. The priorities need to be re-assessed. He emphasised on the need to build indigenous capabilities in the domestic set up even if this is costly and also have global quality OEM and product players.

Today several leading academic institutions in India are doing research on semiconductor. However, the academic work can exist in silos, though they collaborate on need basis. This makes the outcome of collaborations incremental, primarily aimed for showcasing, and difficult to translate this into ground reality or world-class products. For true translation to happen, there is a need of absolute chain of command and empowerment in a collaborative framework. To stay relevant processes need to be streamlined and timelines need to be cut down. All the institutions have their own processes, systems, reporting and they have to deliver something which is functional, transnational in a time bound manner. Thus, for bringing India on the semiconductor manufacturing global map, India needs its own semiconductor Manhattan projects.

It also needs to make a roadmap for being top computing power in the next 10 years. Nevertheless, there are good things happening such as Assembly, Testing, Marking and Packaging (ATMP) proposals to invest in India from global semi giants (such as Micron) for building testing and packaging facilities. Tata has also announced its semiconductor plans. But these are few steps, and more need to be done overall in the national context. A lot of semiconductor designing happening in India, but those are for global tech companies headquartered outside India with almost negligible Indian OEMs. This is painful and disappointing to see. Thus, India must develop its capacity and capability in this field to meet its own requirement and also become an integral part of the global semiconductor value chains. Given that there is no shortage of knowledge, skill, vision and strategy on semiconductor technology in India and it is set for a bright future for the country’s semiconductor aspirations.

(The paper is the author’s individual scholastic articulation. The author certifies that the article/paper is original in content, unpublished and it has not been submitted for publication/web upload elsewhere, and that the facts and figures quoted are duly referenced, as needed, and are believed to be correct). (The paper does not necessarily represent the organisational stance... More >>