What is an ASIC?
ASICs are the fundamental building blocks of modern electronics – your laptop or PC, TV, cell phone, digital watch, almost everything you use on a day-to-day basis. It is also the fundamental building block upon which the FPGA we will be looking at is built from. In short, an ASIC is a custom-built chip designed using the same language and methods we will be introducing in this book.
FPGAs came about as the technology to create ASICs followed Moore's law (Gordon E. Moore, Cramming more components onto integrated circuits, Electronics, Volume 38, Number 8 (https://newsroom.intel.com/wp-content/uploads/sites/11/2018/05/moores-law-electronics.pdf) – the idea that the number of transistors on a chip doubles every 2 years. This has allowed for both very cheap electronics in the case of mass-produced items containing ASICs and led to the proliferation of lower cost FPGAs.
Why an ASIC or FPGA?
ASICs can be an inexpensive part when manufactured in high volumes. You can buy a disposable calculator, a flash drive for pennies per gigabyte, an inexpensive cell phone; they are all powered by at least one ASIC. ASICs are also sometimes a necessity when speed is of the utmost importance, or the amount of logic that is needed is very large. However, in these cases, they are typically only used when cost is not a factor.
We can break down the costs of developing a product based on an ASIC or FPGA into Non-Recurring Engineering (NRE), the one-time cost to develop a chip, and the piece price for every chip, excluding NRE. Ed Sperling states the following in CEO Outlook: It Gets Much Harder From Here, Semiconductor Engineering, June 3, 2019, https://semiengineering.com/ceo-outlook-the-easy-stuff-is-over/:
These costs include more than just the mask sets, the blueprint for the ASIC if you will, that is used to deposit the materials on the silicon wafers that build the chip. It's also the teams of design, implementation, and verification engineers that can number into the hundreds. Usually factored into ASIC costs are re-spins, which are bug fixes. These are a factor because large, complex devices struggle with first-time success.
Compare this to an FPGA. Fairly complex chips can be developed by a single person, or small teams. Most of the NRE has been shouldered by the FPGA vendor in the design of the FPGA chips, which are known good quantities. What little NRE is left is for tools and engineering. Re-spins are free, except for time, since the chip can be reprogrammed without million-dollar mask sets.
The trade-off is the per part cost. High volume ASICs with low complexity, like the one inside a pocket calculator or a digital watch, can cost pennies. CPUs can run into the hundreds or thousands of dollars. FPGAs, even the most inexpensive Spartan-7, start at a few dollars and the largest and fastest can stretch into the tens of thousands of dollars.
Another factor is tool costs. As we will see later in this chapter, Xilinx provides the Vivado tool suite for free in the form of a webpack for the smaller parts. This speeds adoption, where the barrier to entry is now a computer and a development board. Even the cost of developing more expensive parts is only a few thousand dollars if you need to purchase a professional copy of Vivado. ASIC tools can run into the millions of dollars and require years of training since the risk of failure is so high. As we will see in our projects, we'll make mistakes, sometimes to demonstrate a concept, but the cost to fix it will only be a few minutes of time, mostly to understand why it failed:
Figure 1.1 – Simple ASIC versus FPGA flow
The flow for an ASIC or FPGA is essentially the same. ASIC flows tend to be more linear, in that you have one chance to make a working part. With an FPGA, things such as simulation can become an option, although strongly suggested for complex designs. One difference is that the lab debug stage can also act as a form of simulation by using ChipScope, or similar on-chip debugging techniques, to monitor internal signals for debugging. The main difference is that each iteration through the steps costs only time in an FPGA flow. In this situation, any changes to a fabricated ASIC design requires some number of new mask sets that can run into the millions of dollars.
We've briefly looked at what an ASIC is and why we might choose an ASIC or an FPGA for a given application. Now, let's take a look at how an FPGA is created using an ASIC process.