A Brief History of ASICs-vlsi-technology-chip-min.jpgIn the early 1980s the ideas and infrastructure for what would eventually be called ASIC started to come together. Semiconductor technology had reached the point that a useful number of transistors could be put onto a chip. But unlike earlier, when a chip only held a few transistors and thus could be used to create basic generic building blocks useful for everyone, each customer wanted something different on their chip. This was something the traditional semiconductor companies were not equipped to provide. Their model was to imagine what the market needed, create it, manufacture it and then sell it on the open market to multiple customers.

The other problem was that semiconductor companies knew lots about semiconductors, obviously, but didnít have system knowledge. The system companies, on the other hand, knew just what they wanted to build but didnít have enough semiconductor knowledge to create their own chips and didnít really have a means to manufacture those chips even if they did. What was required was a new way of doing chip design. The early part of the design, which was system knowledge heavy, would be done by the system companies. And the later part of the design, which was semiconductor knowledge heavy, would be done by the company that was going to manufacture it.

Two companies in particular, VLSI Technology and LSI Logic, pioneered this. They were startup companies with their own fabs (try getting that funded today) with a business model that they would, largely, build other peopleís chips. In the early days, since there werenít really any chips out there to be built, they had to find other revenue streams. VLSI Technology, for example, made almost all of its money building read-only-memories (ROMs) that went into the cartridges for the first generation of video game consoles.

Daisy, Mentor and Valid, who had electronic design systems mostly targeted at printed circuit boards, realized that they could use those systems for the front end of ASIC design too.

ASIC design typically worked like this. A system company, typically someone building an add-on board for the PC market since that was the big driver of electronics in that era, would come up with some idea for a chip. They would negotiate with several ASIC companies to decide which to select, although it was always a slightly odd negotiation since only a vague idea of the size of the design was available at that point. They would then partner with one ASIC company such as LSI Logic who would supply them with a library of basic building blocks called cells.

The system company would use a software tool called a schematic editor to create the design, picking the cells they wanted from the library and deciding how they should be connected up. The output from this process is called a netlist, essentially a list of cells and connections.

Just like writing software or writing a book, the first draft of the design would be full of errors. But with semiconductor technology it isnít possible to build the part and see what the errors are. Even back then it cost tens of thousands of dollars and a couple of months to manufacture the first chip, known as the prototype. Also, unlike with writing a book, itís not possible to simply proofread it and inspect the schematic, too many errors would still slip through.

Instead, a program called a simulator was used. A flight simulator tells a pilot what would happen if he or she moves the controls a certain way, and, unlike in the real world doesnít cost a fortune if the plane crashes. In the same way, a simulation of the design checked how it behaved given certain inputs without requiring the expense of building the chip. Errors that were detected could be fixed and the simulation run again.

When finally the design was determined to be correct, the netlist was sent from the system company to the ASIC company for the next step. Using a program known as place & route tool the netlist would be converted to a full chip design. In addition to creating the actual layout for manufacture, this process also created detailed timing, just how long every signal would take to change. This detailed timing would be sent back to the system company so that they could do a final check in the simulator to make sure that everything worked with the real timing versus the estimated timing that was used earlier.

At that point the system company took a deep breath and gave the go ahead to spend the money to make the prototypes. This was (and is) called tapeout.

The ASIC company then had the masks manufactured that were be needed to run the design through their fab. In fact there were two main ASIC technologies, called gate-array and cell-based. In a gate-array design, only the interconnect masks were required since the transistors themselves were pre-fabricated on a gate-array base. In cell-based design, all the masks were required since every design was completely different. The attraction of the gate-array approach was that, in return for giving up some flexibility, the manufacturing was faster and cheaper. Faster, because only the interconnect layers needed to be manufactured after tapeout. Cheaper, because the gate-array bases themselves were mass produced in higher volume than any individual design.

A couple of months later the prototypes would have been manufactured and samples shipped back to the system company. These parts typically would then have been incorporated into complete systems and those systems tested. For example, if the chip went into an add-in board for a PC then a few boards would have been manufactured, put into a PC and checked for correct operation.

At the end of that process, the system company took another deep breath and placed an order with the ASIC company for volume manufacturing, ordering thousands or possibly even millions, of chips. They would receive these a few months later, build them into their own products, and ship those products to market.

A Brief History of ASICs-lsi-logic-hq-min.jpgAll semiconductor companies were caught up in ASIC in some way or another because of the basic economics. Semiconductor technology allowed medium sized designs to be done, and medium sized designs were pretty much all different. The technology didnít yet allow whole systems to be put on a single chip. So semiconductor companies couldnít survive just supplying basic building block chips any more since these were largely being superseded by ASIC chips. But they couldnít build whole systems like a PC, a TV or a CD player since the semiconductor technology would not allow that level of integration. So most semiconductor companies, especially the Japanese and even Intel, started ASIC business lines and the market became very competitive.

ASIC turned out to be a difficult business to make money in. The system company owned the specialized knowledge of what was in the chip, so the semiconductor company could not price to value. Plus the system company knew the size of the chip and thus roughly what it should have cost with a reasonable markup. The money turned out to be in the largest, most difficult designs. Most ASIC companies could not execute a design like that successfully so that market was a lot less competitive. The specialized ASIC companies that could, primarily VLSI Technology and LSI Logic again, could charge premium pricing based on their track record of bringing in the most challenging designs on schedule. If you are building a sky-scraper you donít go with a company that has only built houses.

As a result of this, and of getting a better understanding of just how unprofitable low-volume designs were, everyone realized that there were less than a hundred designs a year being done that were really worth winning. It became a race for those hundred sockets.

Semiconductor technology continued to get more powerful and it became possible to build whole systems (or large parts of them) on a single integrated circuit. These were known as systems-on-chip or SoCs. The ASIC companies all started to build whole systems such as chipsets for PCs or for cell-phones alongside their ASIC businesses which were more focused on just those hundred designs that were worth winning. So all semiconductor companies started to look the same, with lines of standard products and, often, an ASIC product line too.

One important aspect of the ASIC model was that the tooling, the jargon word for the masks, belonged to the ASIC company. This meant that the only company that could manufacture the design was that ASIC company. Even if another semiconductor company offered them a great deal, they couldnít just hand over the masks and take it. This would become important in the next phase of what ASIC would morph into.

ASIC companies charged a big premium over the raw cost of the silicon that they shipped to their customers. ASIC required a network of design centers all over the world staffed with some of the best designers available, obviously an expensive proposition. Customers started to resent paying this premium, especially on very high volume designs. They knew they could get cheaper silicon elsewhere but that meant starting the design all over again with the new semiconductor supplier.

A Brief History of ASICs-tsmc-fab-12-min.jpgAlso, by then, two other things had changed. Foundries such as TSMC had come into existence. And knowledge about how to do physical design was much more widespread and, at least partially, encapsulated in software tools available from the EDA industry. This meant that there was a new route to silicon for the system companies, namely ignore the ASIC companies, do the entire design including the semiconductor-knowledge-heavy physical design, and then get a foundry like TSMC to manufacture it. This was known as customer-owned-tooling or COT, since the system company as opposed to the ASIC company or the foundry, owned the whole design. If one foundry gave poor pricing the system company could transfer the design to a different manufacturer.

However, the COT approach was not without its challenges. Doing physical design of a chip is not straightforward. Many companies found that the premium that they were paying ASIC companies for the expertise in their design centers wasnít for nothing, and they struggled to complete designs on their own without that support. As a result, companies were created to supply that support, known as design services companies.

Design service companies played the role that the ASIC companiesí design centers did, providing specialized semiconductor design knowledge to complement the system companiesí knowledge. In some cases they would do the entire design, known as turnkey design. More often they would do all or some of the physical design and, often, manage the interface with the foundry to oversee the manufacturing process, another area where system companies lacked experience.

One company in particular, eSilicon, operates with a business model identical to the old ASIC companies except in one respect. It has no fab. It actually builds all of the customersí products in one of the foundries (primarily TSMC).

Another change has been the growth of field-programmable gate-arrays (FPGAs) which are used for many of the same purposes as ASIC used to be.

So that is the ASIC landscape today. There is very limited ASIC business conducted by a few semiconductor companies. There are design services companies and virtual ASIC companies like eSilicon. There are no pure-play ASIC companies. A lot of what used to be ASIC has migrated to FPGA.