The Rise, Evolution, and Future Frontiers of VLSI

“The tyranny of numbers” was the name given to the seemingly insurmountable barrier that the electronics industry faced in the mid-1950s. The challenge was simple: increasing computational power required more components — transistors, capacitors, and resistors — which then had to be meticulously wired together by hand. The growing number and complexity of such minute interconnections made it both physically and economically impractical to construct more advanced systems. As a result, technological progress began to stall.

Silicon, when treated with specific techniques, can be used as the foundation for many electronic components. Resistors, capacitors, and transistors could be made directly on a single slice of silicon. This insight would eventually overcome the tyranny of numbers; as it allowed engineers to replace a large number of discrete components by a single silicon die. This marked the birth of the integrated circuit (IC).

By the early 1960s, the first true ICs had begun to enter mass production. These early chips contained just a few tens of bipolar junction transistors (BJTs). This stage of development was known as Small Scale Integration (SSI).

In the years that followed, improvements in fabrication technology and digital logic design enabled engineers to place hundreds of transistors on a single chip; ushering in the era of Medium Scale Integration (MSI). As transistor counts per chip increased to the thousands, Large Scale Integration (LSI) was soon reached.

Another type of transistor had begun to gain popularity by the late 1960s. Known as the Metal-Oxide-Semiconductor Field-Effect transistor, the MOSFET offered lower power consumption and simpler fabrication, making them easier to scale. By 1964, MOSFET chips had reached higher transistor densities and lower costs than bipolar chips. This rapid increase in transistor counts laid the foundation for the Very Large Scale Integration (VLSI) era, where chips began to contain hundreds of thousands to millions of transistors.

The increasing transistor density and the corresponding rise of computing power made it possible to construct complex digital systems on a single chip. Among the earliest and most revolutionary of these systems was the microprocessor. The microprocessor was able to perform the core functions of the central processing unit (CPU) of a computer by virtue of its logic and control circuitry. While the CPUs of earlier computers were built using racks of circuit boards, which contained MSI or LSI circuits (Medium/Large Scale Integration), the advent of VLSI and MOS fabrication enabled engineers to pack much greater computing power into a single chip. This greatly reduced the cost of processing power. Simultaneously, it also increased reliability, as a single chip contained fewer connections which could fail. Advances in microprocessor technology would eventually render most other forms of computer hardware obsolete.

In 1965, Intel co-founder Gordon Moore made an observation that would come to guide the semiconductor industry for the next half century. He stated that the number of transistors on a chip was doubling roughly every two years (18 months, to be precise), which resulted in computing power rising exponentially.

However, this greater computing power still consumed the same amount of power. This observation was explained in a 1974 paper co-authored by Robert Dennard, and came to be known as Dennard scaling. If the size of a transistor is decreased by some factor, then the voltage across it and the current through it would also decrease by the same factor. This, in turn, causes a decrease in the power consumed. Simultaneously, the switching frequency of the transistor also becomes faster.

With the above knowledge, even as the number of transistors per chip doubles roughly every two years, the transistors themselves get smaller and more efficient simultaneously. So it is that computing power grows exponentially with little change in the power consumed by the chip.

These principles guided the industry for decades until Dennard scaling began to break down in the mid-2000s, as leakage became a significant factor in IC design. Further, Moore’s Law is also expected to slow down significantly within the current decade, with some predicting that it will halt completely.

As these guiding principles faded, VLSI design gradually shifted its focus from simply increasing raw transistor counts to more efficient integration. Engineers began to explore methods of packing entire systems onto a single chip, without compromising power or speed. As complexity of design rose, hardware description languages such as Verilog and VHDL were created to help streamline the design process.

Multicore processors emerged in the 2000s to keep pace with the ever-growing demand for greater computing power; and sizes of chips shrunk to the nanometer scale. A variety of VLSI-based architectures emerged, each suited to a unique application: high-performance microprocessors which can be programmed for specific use cases, compact microcontrollers which find great utility in embedded systems, and the highly integrated System-on-Chip (SoC) which combine together multiple complex systems.

While each of the above are related, there are also some key differences:

Microprocessors typically maintain only CPU functionality, and rely on external components for memory storage, input/output and timers. Further, their hardware is largely general purpose, and can be used for a variety of applications, from handheld devices to supercomputers. In contrast, microcontrollers combine the CPU, the memory, and the input/output peripherals, reducing the need for external components. Microcontrollers also consume lesser power and offer real-time control, all while being relatively compact and cost-efficient. This makes microcontrollers exceptionally well-suited for such embedded systems applications as air conditioners, refrigerators, washing machines, and glucose monitors. A System-on-Chip, or SoC, represents an even greater level of integration and combines most, if not all components of a computer system onto a single chip. They include CPU(s), memory and input/output, in addition to optional units such as bluetooth, wi-fi connectivity and GPU(s). SoCs represent the epitome of modern VLSI — billions of transistors across discrete components, all integrated onto a single chip.

As advancements are made, it becomes clear that fundamental constraints of physics and manufacturing are being approached. Such effects as quantum tunneling and current leakage can no longer be neglected. Heat dissipation poses a challenge as transistor counts increase. Achieving the requisite accuracy in doping concentrations via photolithography — the primary process used in the manufacture of ICs — is growing difficult.

This slowdown has prompted a shift in the industry. Alternative methods of packaging are being researched and prototyped. Various approaches to stacking multiple dies together have been proposed, such as the 3D ICs and 2.5D ICs. New materials for building transistors, such as molybdenum disulphide (MoS2), gallium nitride (GaN), and graphene are also being studied. Further, neuromorphic computing, which mimics the human brain and uses artificial neural networks to perform calculations offers a radically new approach to electronics. Similarly, quantum computing leverages quantum mechanical phenomena to perform certain tasks much faster than traditional processors.

VLSI has been crucial in the creation of almost all the processing power in the modern world. As traditional scaling slows, the future of VLSI lies not in stretching physical limits alone, but in exercising new approaches to how we build and integrate computing systems.

Sources:

1959: The Year Everything Changed — https://books.google.com/books?id=KCB1pcHgT68C Ross, Bassett (2002). To the Digital Age: Research Labs, Start-up Companies, and the Rise of MOS Technology, pages 12–28 https://web.archive.org/web/20120926194210/http://www.ieeeghn.org/wiki/index.php/Oral-History:Robert_N._Noyce https://www.electrochem.org/dl/interface/spr/spr13/spr13_p055_061.pdf https://www.computerhistory.org/revolution/digital-logic/12/279 https://news.samsung.com/global/qualcomm-and-samsung-collaborate-on-10nm-process-technology-for-the-latest-snapdragon-835-mobile-processor https://semiengineering.com/knowledge_centers/standards-laws/laws/dennards-law/ https://discrete.co.in/blog/different-types-of-microprocessors-in-embedded-systems Kumar, Suhas (2012). “Fundamental Limits to Moore’s Law” https://insemitech.com/blogs/exploring-the-challenges-of-vlsi-design-navigating-complexity-for-success/