When we talk about the design of new integrated circuits on a chip, whatever they are, we are already talking about a CPU, a GPU, a chipset, a memory, etc. Three different styles or methodologies are used in terms of design. Which are called Full Custom, Semi Custom and EDA. Let’s see what each of them looks like and how they differ from each other.
What is the Standard Cell Library?
First of all, we need to understand what a standard cell library is. In the same way that programming languages generally have libraries with the data structures and algorithms most used when developing programs. foundries or processor factories have their own standard libraries where in each of them are well defined the logic gates that will be used to create a future digital design.
Each manufacturing node has a standard library of logic gates depending on the node and utility to be used. For example, TSMC has a 7nm node for high performance and another for low power. Each of them has a library of standard cells, which will be essential for the creation of new processors. Since it includes everything you need like logic gates, toggles, locks, etc.
It is therefore the foundries themselves such as Intel, TSMC, Samsung, GLOBALFOUNDRIES, SMIC and many others in the world who each end up developing their Standard Cell Library, which will be adopted for the creation of new chips. Each standard cell library can have different variations of each of the logic gates. Some will stand out for their performance, others for their smaller surface, we will have versions of a logic gate with lower energy consumption and also with knowledge.
Fully Custom Chip Design Methodologies
The Full Custom methodology is the most complex of all, since it is only adopted by a few companies in the world with immense research and development capital. This methodology also requires many years of development to achieve a design. It is based on the combination of using the chosen node’s standard cell library and a custom design.
What is? Suppose we have a combination of logic gates that is not built into the standard cell library and whose implementation is critical to your design. Well, companies with the Full Custom feature have the option to extend the standard cell library of the node they are using and create new add-ons to use in their own designs.
Typically, processor design companies have planned for a very long-term processor, the node they will be using is not ready, and their standard cell library is not available. These companies therefore design their own logic gates for a more advanced node. This is something even Fabless like AMD and NVIDIA are doing to test new architectures. The methodology is for foundries to advance their design libraries to them in advance before starting production on the new node.
From them, companies with the Full Custom capability create simple preliminary designs and thus suggest improvements on the future manufacturing node. This is the reason why certain foundries and industrialists seek to have partners with more research and development capacity. The counterpart of this method? Its high cost, not only due to the design of new combinations of logic gates, but also to the fact that they are designed for the development of processors within three to five years.
Semi-custom chip design methodologies
Semi-custom designs are those that only use the standard cell library of a node already available for manufacturing. It is therefore a design methodology designed to launch processors in the very short term and not requiring large research and development capital. Today, over 80% of chip design companies are semi-custom. All designs start out as semi-custom designs, that’s when the more experienced engineers see the ability to create a better processor using their own technology, they go from semi-custom to fully custom. But as we’ve discussed before, it’s something very few companies can afford.
A curious case of a semi-custom design is AMD, which has its own standard cell library using TSMC as its main manufacturer. A company that accepts the design of a chip with AMD will not only be able to use technologies already developed by AMD itself, but will also be able to use its own library of standard cells to create coprocessors such as accelerators and domains. specific processors for its designs. Something SONY and Microsoft have done to create the main processors for their next generation consoles.
However, despite its lower price compared to the Full Custom design, it requires the chip manufacturing to be outsourced to large foundries, which will not accept very small orders. The design of the chips therefore involves enough capital to pay for the manufacture of the chips.
Deployment in FPGA and CPLD
The use of FPGAs has become popular in recent years as well as CPLDs, both of which are programmable logic whose operation and the differences between each one will not be explained here. Its use in design methodologies for new processors occurs in two cases. The first for prototyping, the second occurs because there are times when a company does not have the economic capacity to contract a large volume of chips, which is not in the best interests of foundries.
It must be taken into account that deployment in each manufacturing node is increasingly expensive to implement, so large foundries are looking for the most economically powerful customers. You can design the best GPU in the world that if you don’t have the capital from NVIDIA or AMD to lease a large volume of chips then your design will not see the light of day.
Due to this situation, many chip design companies do not manufacture them, but sell their designs to be implemented in an FPGA or CPLD that they implement directly into their products. The FPGA ends up becoming the chip that should have been mass-produced. In return, there is no need to pay for mass production and implementation is done in no time.
The problem with FPGAs in particular is that the cost per chip is obviously higher and these are solutions that will be very expensive, but are designed for very specific customers. Who have the money to pay for the solution, but no time to wait for a solution.