2
\$\begingroup\$

VLSI and advances in our understanding of semiconductor physics has made it possible to have enormous computing capacity at our fingertips.

However, I never really understood what "integration" means in VLSI. Why do we want integrated circuits and what do we aim to achieve with integration?

Is it about using FETs as capacitors so you get a voltage controlled capacitor and a switch in one physical component? But, we intend to use transistors as switch for conventional computing, right?

Is it about providing an electronic basis to perform Boolean algebra? Is it about having digital logic units using (for instance) pass transistor logic to have more functionality while using less components?

Is it modularized design and building on the top of pre-designed building blocks and separation of design from implementation technology as suggested by Mead-Conway chip design revolution?

Or is it about advances in optical technology so as to be able to carve out ever shrinking feature sizes on a silicon die?

\$\endgroup\$
2
  • 4
    \$\begingroup\$ What are you asking that's not covered in, say, the wikipedia entry for "integrated circuit" or "Invention of the integrated circuit"? \$\endgroup\$
    – jonathanjo
    Sep 19, 2022 at 17:04
  • 2
    \$\begingroup\$ Have you ever tried building a computer without integrated circuits? Even as a thought experiment? \$\endgroup\$
    – user253751
    Sep 19, 2022 at 21:48

5 Answers 5

13
\$\begingroup\$

You’re overthinking this. An ‘integrated circuit’ (IC) just means a device that has more than one active element (diode, FET or BJT) implemented on a single die. That’s it.

The IC is differentiated from discrete components (one active component per die) and from hybrid modules (multiple die on a substrate in a package).

ICs have been around for a long time, first developed in late 1958 (Jack Kilby, TI) and introduced in practical monolithic form in 1959 (Bob Noyce, Fairchild).

Nevertheless it took a while for integrated circuits to catch on. In the early days they were considered exotic, expensive and unreliable. Systems vendors like IBM and their competitors continued to use discretes or hybrids, migrating to ICs later in the 1960s.

Meanwhile, it was NASA that really pushed IC technologies, notably by being the prime customer for first mass-produced IC: a dual 3-input RTL NOR gate. This chip integrated six NPN transistors and eight resistors on one silicon die.

Here is its schematic:

enter image description here

And a die photo:

enter image description here

(From https://en.m.wikipedia.org/wiki/Apollo_Guidance_Computer)

This simple IC was used to construct all of the logic in the Apollo Guidance Computer (AGC), developed by MIT. The entire AGC (minus its DSKY terminal) was housed a single 70-pound box, at a time when computers occupied multiple cabinets in raised-floor rooms. It had the approximate compute power of a 6502.

More about the AGC here: http://www.righto.com/2019/09/a-computer-built-from-nor-gates-inside.html

And yet this simple NOR gate device is an integrated circuit, just like todays largest CPUs and SoCs. It only differs from today’s monster chips in extent of its integration.

Why do we want integrated circuits and what do we aim to achieve with integration?

In the beginning, for the same reasons as NASA sought to achieve:

  • Reduced physical size and weight
  • Greater functional density
  • Design standardization and reuse
  • Economies of scale

Then later, as IC tech became mainstream:

  • Lower cost
  • Lower power
  • Higher performance
  • Expanded application spaces

NASA faced a lot criticism at the time for adopting such an ‘unproven’ technology when it chose MIT’s IC-based approach. Yet, here we are, enjoying the fruits of it some sixty years later.

And to a certain extent we’ve come full circle. The largest ICs (like AMD’s EPYC) are actually multichip modules (MCM) that is, multiple die (‘chiplets’) on a single substrate.

The motivation for chiplets is that the yield loss of large die becomes uneconomic at certain size thresholds. Dividing up larger chips into chiplets mitigates this yield-loss cost, at the expense of a more complex MCM package.

Another emergent technology is 3D chip fabrication, which has come on line as a mainstream memory technology (NAND flash especially). It can also be applied to good advantage for high performance processor technology for networking, DSP and AI and others.

\$\endgroup\$
1
  • 1
    \$\begingroup\$ Could not have had a more direct answer! Thank You! \$\endgroup\$
    – lousycoder
    Sep 20, 2022 at 7:11
5
\$\begingroup\$

Integration just refers to how many elements (transistors, resistors, capacitors, etc.) you can fit onto a single chip. The more transistors on a chip, the higher its integration level.

\$\endgroup\$
5
\$\begingroup\$

The alternative to an integrated circuit is a circuit built from discrete elements.

Each transistor, resistor, capacitor, etc., would have its own package, and be mounted individually on the circuit board.

Why do we want integrated circuits and what do we aim to achieve with integration?

Using an integrated circuit, particularly for a complex circuit, requires vastly less area and power than a similar circuit built from discrete elements.

\$\endgroup\$
3
\$\begingroup\$

It seems about integration which means a gathering of lot of different types of components such as diodes,transistors (MOSTFET mostly), resistors, capacitors etc. VLSI uses logic gates which are often viewed as switches made of transistors. It is about a large logic system with just transistors as switches which turn on and off very fast. The functioning capability of a VLSI circuit such as a IC or ASIC is usually much faster than a MCU in analog or digital signal processing and analog or digital signal generation. It is mainly about performance of hardware for the bucks you pay for it.

\$\endgroup\$
1
\$\begingroup\$

It is an outdated term. Don't worry about the term "VLSI."

But basically, shrinking transistors has allowed us to design processors with more cache which dramatically improves speed and performance. Processors are mostly cache (SRAM). The logic doesn't take up that much space.

But small transistors have allowed miniaturization of pretty much all IC functions except those that dissipate lots of power (e.g., high-power amplification).

\$\endgroup\$
11
  • 2
    \$\begingroup\$ The context of VLSI in the sequence of VLSI, LSI, MSI, SSI is outdated. 'VLSI' came to mean 'lots of transistors on a chip' and today originally what was thought of as VLSI (1980s microprocessors) are tiny compared with today's. The term VLSI got picked up by various things (conferences, journals, societies) that got going in the 1980s, and that way it's stuck around. Nobody would call their new chip today 'VLSI'. \$\endgroup\$ Sep 20, 2022 at 12:57
  • 3
    \$\begingroup\$ @ReversedEngineer: The distinction between VLSI and LSI is pretty much outdated, rendering the "V" a bit quaint. The situation would be somewhat analogous to categorizing bridge spans as "short" if they were less than three meters long, "medium" if they were three to ten meters, "long" if they were 10 to 30, and "very long" if they were over 30 meters. If spans over 30 meters were commonly referred to as "very long", with the term "long" reserved for those in the 10 to 30 meter range, an engineering company that billed itself as specializing in "long" bridge spans... \$\endgroup\$
    – supercat
    Sep 20, 2022 at 15:24
  • 2
    \$\begingroup\$ ...might be seen as specializing only in those which, by modern standards, weren't very long. If, however, a company specialized in spans in the 30 to 100 meter range, such spans normally wouldn't be categorized as "very long". \$\endgroup\$
    – supercat
    Sep 20, 2022 at 15:50
  • 1
    \$\begingroup\$ See also "short wave radio." \$\endgroup\$
    – user57037
    Sep 20, 2022 at 18:45
  • 1
    \$\begingroup\$ @ReversedEngineer maybe things have changed a bit now that we have multi-core processors with on-die GPUs. Cache is pretty important to performance. But maybe it is no longer > 50 percent of the die area. \$\endgroup\$
    – user57037
    Sep 21, 2022 at 16:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.