Sand plays a vital role in our daily lives. Used in its raw form, it is the foundation material for building homes. Purify the sand a little more and it becomes the foundation of the semiconductor industry.
India is currently waking up to its opportunities vis-à-vis semiconductors: access to the underlying technologies has been a long-standing dream of our nation. Success on this front would place India among a small, elite group of nations that have access to the tech as well as provide thousands of highly skilled jobs.
In a major setback, however, Foxconn Technology Group recently withdrew its support from its joint venture with Vedanta, Ltd. to establish a semiconductor manufacturing plant in Gujarat. The Indian government has also introduced incentives to catalyse this sector but only time will tell if they will bear fruit.
Initially, the plan was to establish a manufacturing unit for a 40-nm node. After Foxconn’s withdrawal, Vedanta has maintained that it acquired the relevant technologies from another major company. It is also in the process of acquiring the technologies for the 28-nm, 63-nm, and 90-nm nodes.
As a result, the term “semiconductor node” has become the talk of the town. What does it mean exactly?
What is a semiconductor chip?
At its core, a semiconductor chip is composed of transistors, which in turn are meticulously crafted from a specially selected material, typically silicon. One major function of a transistor is to encode information in the form of 0s and 1s, and to manipulate them to produce new information.
These transistors have three parts: the source, the gate, and the drain (or the sink).
The flow of current between the source and the drain points is regulated by the voltage applied to the gate. This arrangement gave rise to the specific meaning of ‘gate’ in computing – analogous to a physical gate, but operating with electrical means rather than mechanical ones.
By manipulating the gate to ‘open’ or ‘close’, the transistor stores and manipulates the data in a semiconductor chip. The semiconductor stores information in the form of bits. Each bit is a logical state that can have one of two values (represented by voltage levels) at a time. The more bits a semiconductor can store and the more quickly it can manipulate them, the more data transistors can process.
The three parts of a transistor are connected to multiple metal layers on top of them that apply voltages, forming a complex mesh of electrical connections with the transistors. The metal layers allow selective access to a transistor and provide the versatility required for the chip to execute multiple tasks.
What does the node number mean?
Through history, the names of semiconductor nodes have been based on two numbers: the length of the gate and the distance between adjacent metal strips connected to the gate; the latter, when measured centre to centre, is called the pitch. These dimensions were often equal. For example, the 500-nm node featured a gate length and metal half-pitch of 500 nm. This naming convention started with the invention of the transistor in 1960, with the 50,000 nm (50 microns) node, up to the 350-nm node of the 1990s.
The size of transistors has progressively shrunk over the years. The smaller a transistor becomes, the more of them can be fit on a semiconductor chip, the more data the chip can store, the more computing power there will be. For a sense of scale: in the early 1970s, the transistor density per sq. mm on a chip was around 200 – whereas a chip within an iPhone has around 100 million transistors per squared millimetre. This is the incredible progress researchers and engineers have made in the last half-century.
Yet as transistors continued to become smaller, researchers spotted a discrepancy between the gate length and the metal pitch, rooted in the fact that while smaller transistors generally resulted in faster operation, reducing the size of metal wires created different problems, including not being able to transport data fast enough.
In 1997, a 250-nm semiconductor node hit the market – and also broke the naming convention. Its metal half-pitch contributed to the name, but its gate length, which had been reduced to 200 nm, didn’t. Since then, as the miniaturisation continued, both the half-pitch and gate length ceased to contribute to the node name.
For example, in the 130-nm node, the half-pitch measured 150 nm while the gate length measured 65 nm. Today’s state-of-the-art 7 nm node in fact has no physical parameters that come close to 7 nm. This is because it’s impossible to reliably fabricate features around 7 nm with existing technologies.
From a technical standpoint, node names hold no significance vis-à-vis the actual physical dimensions. Instead, marketers use them to mean one node is better than a previous iteration.
In fact, different companies have also been using “nm” in the name to mean different things. Intel’s 10-nm node and Global Foundry’s 7-nm node have similar gate length (around 54 nm), metal pitch (around 40 nm), and working efficiency. The only information that can be derived from the node number of a particular company is that it is an improvement on its predecessor.
Does India need legacy nodes?
The choice of nodes, just like our choices in life, involve compromises. While advanced nodes range from 10 nm to 5 nm, India’s current focus is around 28 nm or higher. However, this doesn’t mean we are attempting to develop outdated chips. Starting with legacy nodes can offer numerous advantages, including equipping us for long-term success.
While the most advanced nodes are used in devices like smartphones and laptops, many applications require legacy nodes, including robotics, defence, aerospace, industry automation tools, automobiles, Internet of Things, and image sensors – because they are more cost-effective.
The principal revenue source for any fabrication facility, or ‘fab’, is its most advanced node. But almost every commercial fab also maintains the production of legacy nodes to fulfil demands in the aforementioned areas. For example, in 2022, half of the revenue of Taiwan Semiconductors was from 5 and 7 nm. The other half was from 16-250-nm nodes. The revenue from 40 nm alone accounted for 8%.
Indeed, as the demand for electric cars – together with the ever-increasing demand for complementary electronics in the car, like music players – increases, the demand for legacy nodes will also increase. According to an October 2022 report by McKinsey, the value of semiconductor technologies in automobiles in the “user experience and infotainment” category could increase from $11 billion in 2019 to $30 billion in 2030.
Given these facts, the Indian government and private players are sensible to begin their semiconductor journey with the legacy nodes, improving their game over time. Who knows – maybe one day India will be the semiconductors hub of the world.
Awanish is a senior fellow at CERN and has worked at an associated lab of IMEC, a semiconductor foundry, in Belgium.
