Industry Expert Blogs

I want to take a few moments of your time to talk about the business of design/silicon IP. If I look at the EDA industry today (not talking of ARMs and Imaginations of the world), customers usually buy IPs from the big 2 eda vendors, or one of the smaller players or, if they are a FPGA user, may contact a company listed in the FPGA vendor's catalog.

The challenge with buying IPs from a FPGA catalog are typlically the quality of the IP (buggy?) and the level of support or lack thereof. If we look past these 2 issues for the moment,

There are primarily 2 parts to a design IP:

a. digital controller

b. Analog PHY

The big 2 EDA vendors are probably the only ones who have both (a) and (b) for interface protocols of interest, whereas most small vendors have either (a) or (b) and have a partner for the piece they don't have.

Considering many customers prefer to deal with one vendor, and not deal with issues arising out of integrating (a) and (b), tend to go with big 2. Obviously, there is a perception of lowered risk also if a company chooses the big EDA vendors as their IP supplier. If companies do this, there is a large price mismatch as big EDA and some small companies license on a per project basis; while most other smaller companies give an architectural license for a one time fee.

When an IP vendor goes to a any potential customer, the first few questions they are asked is:

1. Is your IP production proven?

2. How many tapeouts has it seen?

3. Any silicon failures?

4. How many customers do you have?

Companies that have convincing answers to these questions are able to charge a premium and sell single use licenses, whereas companies that dont have answers to all these questions, either lose out or end up selling a perpetual architectural license for a throw away price.

This essentially implies that there is significant cost, other than design cost, a company needs to incur to get into the IP business. This includes:

1. Should build a digital controller and a PHY

2. Spend money to get a testchip done on a MPW

3. Target the PHY for a process node majority of customers may need, to save design re-work costs

4. Give away the design IP To first few customers to build a base so you can charge premium

Obviously, by the time a small player does #1-4 for a given IP, the amount of time and cash flow it will take, many other players in the market will be ready with the same IP!! Essentially its the perennial chicken and egg situation!

So what are the possible ways a startup can establish a toe-hold in the IP business? This the million dollar question and there is no one answer except wishing these new kids on the block good luck and Bon Voyage!!

Shishir Gupta