There is a lot of confusion with regard to devices like ASICs, ASSPs, SoCs, and FPGAs. Is an SoC an ASIC, or vice versa, for example?
I often receive questions about the differences between various types of devices, such as ASICs, ASSPs, SoCs, and FPGAs. Is an SoC an ASIC, or vice versa, for example? What's the difference between an ASIC and an ASSP? And should a high-end FPGA be classed as a form of SoC?
There are several problems here, not the least that the technologies and terminologies have evolved over time. Keeping this in mind, the following is my highly simplified interpretation of where these terms came from and what they mean today.
Let's start with an application-specific integrated circuit (ASIC). As the name suggests, this is a device that is created with a specific purpose in mind. When most people hear the term ASIC, their "knee-jerk" reaction is to assume a digital device. In reality, any chip that is custom-made is an ASIC, irrespective of whether it is analog, digital, or a mix of both. For the purposes of these discussions, however, we shall assume a chip that is either wholly or predominantly digital in nature, with any analog and mixed-signal functions being along the lines of physical interfaces (PHYs) or phase-locked loops (PLLs).
ASICs are typically designed and used by a single company in a specific system. They are incredibly expensive, time-consuming, and resource-intensive to develop, but they do offer extremely high performance coupled with low power consumption.
Application-specific standard parts (ASSPs) are designed and implemented in exactly the same way as ASICs. This is not surprising, because they are essentially the same thing. The only difference is that an ASSP is a more general-purpose device that is intended for use by multiple system design houses. For example, a standalone USB interface chip would be classed as an ASSP.
A System-on-Chip (SoC) is a silicon chip that contains one or more processor cores -- microprocessors (MPUs) and/or microcontrollers (MCUs) and/or digital signal processors (DSPs) -- along with on-chip memory, hardware accelerator functions, peripheral functions, and (potentially) all sorts of other "stuff." One way to look at this is that if an ASIC contains one or more processor cores then it's an SoC. Similarly, if an ASSP contains one or more processor cores then it's an SoC.
On this basis, we could view ASIC (and ASSP) as being the superset term because it embraces SoC, or we could regard the SoC as being the superset term because it includes everything in an ASIC (or ASSP) along with one or more processor cores. Are we having fun yet?
ASICs, ASSPs, and SoCs offer high-performance and low power consumption, but any algorithms they contain -- apart from those that are executed in software on internal processor cores -- are “frozen in silicon.” And so we come to field-programmable gate arrays (FPGAs). The architecture of early FPGA devices was relatively simple -- just an array of programmable blocks linked by programmable interconnect.
The great thing about an FPGA is that we can configure its programmable fabric to implement any combination of digital functions we desire. Also, we can implement algorithms in a massively parallel fashion, which means we can perform a humongous amount of data processing very quickly and efficiently.
Over time, the capabilities (capacity and performance) of FPGAs increased dramatically. For example, a modern FPGA might contain thousands of adders, multipliers, and digital signal processing (DSP) functions; megabits of on-chip memory, large numbers of high-speed serial interconnect (SERDES) transceiver blocks, and a host of other functions.
The problem is that the field-programmable gate array (FPGA) moniker no longer reflects the capabilities and functionality of today's programmable devices. We really need to come up with some new terminology that embraces everything today's state-of-the-art tools and technologies are capable of doing.
Of particular relevance to our discussions here is the fact that today's FPGAs can contain one or more soft and/or hard core processors. On this basis, should we class this type of FPGA as being an SoC? Well, personally I have to say that SoC doesn’t work for me, because I equate the term "SoC" with a custom device created using ASIC technology.
Another alternative would be to call these devices Programmable SoCs, or PSoCs, but Cypress Semiconductor has already got the PSoC moniker locked down. The Cypress devices feature a hard microcontroller core augmented with some programmable analog and programmable digital fabric (the digital fabric is more CPLD than FPGA).
Altera used to call its versions of these devices -- the ones that combine hard MCU cores with programmable FPGA fabric -- SoC FPGAs, but they seem to have evolved to just calling them SoCs. Meanwhile, Xilinx calls its flavor of these devices "All Programmable SoCs."
Personally, I'm undecided as to what would be the best name. I think I'd prefer to use PSoC if the folks at Cypress hadn’t already taken control of this appellation, but they have, so we can't. Failing this, I guess I'd opt for SoC FPGA... unless you can suggest something better.
— Max Maxfield, Editor of All Things Fun & Interesting