Articles
Interview with a Fabless Industry Veteran: Ron Jones
Ron Jones is a recognized expert on the outsourced fabless semiconductor environment. CEO of N-Able Group International , Jones spearheads the Business Management Systems Subcommittee of the Fabless Semiconductor Association (FSA), which strives to raise awareness of how management systems can increase the effectiveness of fabless companies. Recently, Tensoft’s Director of Business Development, Caprice Murray, had an opportunity to interview Ron, and discuss fabless industry history and his vision for the future.
MURRAY: Can you talk a little bit about your career to date in this industry?
JONES: I’m a career semiconductor guy. I went into the industry directly out of college — going into R&D initially, and then moving to operations for the next 20 years, most of it at Texas Instruments (TI). I first worked in, and then ran wafer fabs and assembly operations. So, I have a very heavy operations background.
From 1985 to 1995, I worked with three different assembly test subcontract companies, one in Korea, one in the Central Valley of California, and one in Thailand. During that time, I got an appreciation for all the different types of information that customers wanted out of their manufacturing partners. So in 1996, I started N-Able Group, which is a semiconductor-focused consulting and executive search firm. Over the past 11 years, we’ve done a number of projects, advising companies on the operational side, but also fabless companies on selection of software and software companies on how they can better tune up their offerings to satisfy the needs of fabless companies.
Our very first customers were TSMC and Macronix, both Taiwanese companies. We were actually subject matter experts, going in with EDS. The project at each company was the selection of an ERP system and selection of an advanced planning/scheduling system. So very early in that project, it became apparent that there was one big issue: You could have the best software in the world, but if you couldn’t get the data out of the supply chain to feed it, it was problematic.
So these first two projects are what got us started working on the Semiconductor Manufacturing Data eXchange (SMDX) data standard. The companies that had the biggest problem trying to get data out of the supply chain were the fabless guys, because they had a totally outsourced supply chain.
We began talking to the FSA (Fabless Semiconductor Assoc.) about this, and the SMDX committee — later renamed the Business Management Systems Committee — was formed as a result. I’ve chaired that committee since its inception 10 years ago. Each year, we hold a series of forums for ERP software, fabless supply chain management, yield management, etc. software. All the major vendors in the space come in and give an hour-long presentation, so fabless companies and other companies can get an idea of what’s available. Doing this keeps us pretty much on top of what’s going on in the space.
MURRAY: How has the fabless industry changed since you first started out?
JONES: N-Able Group joined the FSA in 1996, their second year. Back then, there were about 50 fabless companies, heavily centered in California’s Silicon Valley. Now, there are literally hundreds of fabless companies — not just in Silicon Valley — but all over the world. Places that you wouldn’t expect. There are 25 to 30 companies in Ottawa; there are companies in Colorado, Texas, and so on. Beyond that, there are all types of fabless companies in Asia, including China, Taiwan, the Philippines, etc. So the industry has grown very significantly over the past 12 years.
The other thing that’s happened over this same time period, is many of the companies that used to do virtually all of their own manufacturing (IDMs or Integrated Device Manufacturers), have changed to a ‘Fab Lite’ strategy. They still have some of their own wafer fabs, they still have some of their own assembly factories, but more and more, they’re outsourcing a lot of their manufacturing to the same kinds of companies that the fabless companies use. Because of this shift, these big firms are beginning to experience the same kind of problems that fabless companies do, of being able to get the data in and manage this extended supply chain.
MURRAY: What about changes in the last few years?
JONES: Some of the biggest changes we’ve seen over the past few years are in the data itself. If you think about it, if you’ve got systems, then you’ve got to have information — whether it’s punched in or electronically transferred or whatever. And if you go back to circa 1995 or 1996, most of the information coming out of the semiconductor supply chain — at least from the fabless side — came in the form of faxes.
Every day, the assembly subcontractors would fax over a work in process (WIP) report or a yield report, and the planner or engineer would sit down and look at it. If there was some analysis they needed, they would have to sit down and key it all into an Excel spreadsheet and work with it. The next day, they’d be almost back to square one again because they’d receive a new fax, and they didn’t know what had changed, etc. It was an almost totally manual environment.
Now, most of this data is available in electronic format. You can get it electronically as an Excel spreadsheet or an XML-tagged file. You can have it sent via email. You can download it off a website. It can be FTP’d to you. Today, there are a variety of ways to get the data in electronic format. As I said… that’s the good news.
The bad news is there are still no data standards. When you get data from TSMC or UMC or Chartered, each one is going to come with a different data format. Standards that were hoped to have a big impact, like RosettaNet, have never really taken off to a significant degree. We’re still pretty much in this point-to-point type of an environment. The key in all of this is to be able to get the data. You can have the greatest systems in the world — the absolute best systems — but if you can’t get the data in an effective fashion, then it’s a little like having a very expensive race car and no gasoline. It doesn’t do any good unless you’ve got the data.
Fabless manufacturing is just a form of semiconductor manufacturing. It’s an outsourced variety, but all of the operations are the same. You still have fab, probe, assembly test, etc. The big difference, however, is between semiconductor manufacturing and what’s called “discrete” manufacturing, which is the type of manufacturing that most of the world’s manufacturers do. What discrete means is taking a lot of components and putting them onto a board. Taking multiple boards and putting those together into a box, then into multiple boxes, etc. Pretty soon, you’ve got a 747 or a BMW. The discrete environment is ‘many-to-one’ — many, many components going into one finished component or product.
Semiconductors on the other hand — whether it’s fabless or a regular IDM semiconductor company — has what’s called an ‘inverted bill of materials.’ Kind of a funny term, but it basically means a ‘one-to-many’ bill of materials. So you take one semiconductor wafer and a mask set, and you put those together into a semiconductor process and after 200 or 300 or 400 steps, you come out with one or many different I.C. chips. Each of those chips can then be tested to different criteria, and so each chip can be ‘binned out’ into different devices. Each of those binned chips can then be put into a multiplicity of different packages. So you have package/chip combinations and then each package/chip combination can be tested to different criteria — either speed criteria or prime and industrial, which is a way of saying military or commercial. So when you look at the overall scheme of things, you’ve got one wafer and mask set that can go into literally dozens, scores or hundreds of products.
As simple as that sounds, it’s a major issue for a lot of software, because the algorithms are different. It’s not that one is harder or easier, because a lot of people say ‘I just can’t imagine how you guys can build semiconductors. Everything is so tiny, etc.’ Well, based on my experience, I sort of have the opposite opinion. If I was running an operation where I had to have a thousand components in the same place at the same time to be able to ship one product, it would drive me crazy! So, I guess it’s all in what you’re used to.
Anyway, the key in this whole thing is to be able to satisfy these special requirements of semiconductor companies and the inverted bill of materials. And I think as we go through, we’ll give some additional examples.
MURRAY: What are the top 3-5 hardest things for an ERP system to do?
JONES: ERP software is broken into two pieces. First, there’s the financial side (accounts payable/receivable, general ledger, those types of things). ERP software does a good job for most semiconductor operations on the financial side. Those are general financial operations. The exception is unit cost that ties to the inverted bill-of-materials. The rub comes in the manufacturing side of it — the master scheduler and things like that. The algorithms that are used for discrete manufacturing to calculate, for instance, available-to-promise or unit cost are totally different. So you can stick stuff in, but you’re not going to get the right answers, and in most cases, it won’t even work.
So, over time, people have developed software to essentially fill in — in place of the ERP manufacturing modules — to take care of the things that semiconductor operations need. The whole thing is basically built around the inverted bill-of-materials, asking: “What processes did the lot go through?” In other words, you have a wafer fab lot, which can be split in two with five wafers going one place and five wafers going to another. You can take it and saw it up. You can put it into three or four different packages. Each package/die combination can be split across multiple manufacturing machines in the manufacturing operation, and then combined back together. Things can be scrapped out and bonused in. All of these things are virtually impossible to track in a discrete manufacturing module.
I mentioned a little earlier the things like ‘available-to-promise.’ The algorithms are totally different to say ‘OK, when can you ship this many parts?’ In order to answer this question, there are calculations that must be done, based on forward-looking projections such as ‘this operation is going to beget this many parts, etc.’ and those don’t work using traditional ERP software. ‘Unit cost’ is the most famous one. You’d think, ‘Well, if the finance system works, (if I can pay bills, general ledger, things like that), surely I can calculate product cost. Well, the algorithms that are used for discrete manufacturing are basically that you add up costs from the bottom, all the way to the top. In semiconductors, you start at the end and you start back-calculating, because you have yields. Not every part that goes in ends up out the back end.
These are examples of where traditional ERP systems break down. So what happens typically today, is companies will buy an ERP system and they’ll use the financial systems and maybe the sales order systems, but for the manufacturing tracking part of it, they’ll buy another piece of software that operates in concert with the ERP financial software to provide a total solution. The hardest thing for an ERP system is just being able to manage the weaknesses in the manufacturing module — things such as unit cost, available-to-promise, lot traceability and lot genealogy.
If you’re a fabless company and your customer comes to you saying ‘You shipped me a defective product.’ You want to be able to track that product all the way back through the manufacturing process to know exactly where it was tested, where it was assembled, and what parts were used to assemble it. How was it probed? How was it manufactured in the fab process? Now, you may say “That’s not the fabless company’s responsibility.” But the customer is looking to the guy who sold him the product to be responsible for providing that information. Just saying “Well, I think it’s this” isn’t good enough, because the customer will say “OK. Then don’t ship me any more unless you can tell me exactly what happened, and assure me that it can’t happen again.” So at the end of the day, the problems for traditional ERP systems really revolve around the manufacturing module and its inability to handle process flow manufacturing.
MURRAY: You’ve advised many fabless companies on their business management software selection — what are your key recommendations for what fabless companies should look for in a vendor?
JONES: What I’ll do first is address the vendors themselves, because you can have instances where the software is very good, but the vendor isn’t, or the vendor’s good but the software is inadequate, and then obviously all the combinations in between. So if we try first to separate the software from the vendors, the key criteria we usually key off when we recommend choosing a software provider is that the company really understand the inverted bill-of-materials. In many cases we see examples where the sales guy says “Sure. Our software will do that” when it really doesn’t. What’s needed is a company that completely understands the nuances of semiconductor manufacturing. Nobody has to know how to build a microprocessor, but they need to understand the inverted bill-of-materials and how it impacts available-to-promise and unit costs, etc.
A second thing that I’d say that’s very important in choosing a vendor is finding one that is financially viable — particularly if that vendor is supplying a stand-alone piece of software. Typically in these cases, a vendor will install an ERP system and their own fabless package as an integrated duo. If that duo goes in and two years later, you need upgrades and the company’s out of business, you’re in big trouble.
Another thing that I believe is key to picking the right vendor, is one that has adequate resources. This may sound obvious, but this isn’t a huge industry. There are probably a dozen or two implementations per year, so if you sign up to have someone put in a fabless package, you need to make sure they have the resources to do it — for a number of reasons. First of all, the longer it takes, the greater the chance that it won’t go in successfully. That may sound counter-intuitive, but our experience is the faster and more focused the implementation, the better the results. And in this business, things always change. That’s another reason to get it done as quickly as possible. It’s costly. Even if someone gives you the software, it’s still going to cost you money because your organization has to spend time and bandwidth — taking themselves away from other activities — to put it in.
So in summary: make sure the company has knowledgeable resources to be able to come in, assess what needs to be done, put it in, tie a ribbon around it and get out of your hair. Then, as I mention before, make sure they’re going to be around, so when you need an upgrade six months or a year down the road, they’ll still be there to help you.
MURRAY: You’ve seen a lot of fabless industry software vendors come and go over the years — what are some of the characteristics of the successful ones?
JONES: The most successful attribute is this understanding of the inverted bill-of-materials. If a company doesn’t have an understanding of that, you’re dead before you even start. They need to understand about semiconductors — how the process flows, the data that’s available and how to manipulate that data to get the right answers for semiconductor process flow manufacturing.
A second thing that you see is that the financial side of an ERP package works properly. The most successful fabless software vendors are ones that not only understand the inverted bill-of-materials, but also understand the “universe” of what a fabless company needs.
The most successful companies we’ve seen have stood back, looked at the big picture and said “Here’s what’s needed. Here’s what ERP provides. And here’s the gap.” The software can be quickly installed, it’s usable and it supports the customer’s need to track both their manufacturing and their financials.
MURRAY: On the flip side of that question, what characteristics have killed off the unsuccessful ones?
JONES: Software vendors that don’t make it have a number of things in common. The ones that die typically don’t understand the inverted bill-of-materials for semiconductors. They think it’s just a sort of a slight variation on discrete manufacturing, and that dooms them to failure. Or, a company will start writing its software, and it takes money from it’s first customer to fund development, but then ends up with software that meets the needs of that first customer but is only 10% reusable for the next customer.
A software company needs to hold itself up and make sure it’s satisfying a broad set of common industry issues. Now, I’m the last person to say you shouldn’t make your customer happy. If the customer want a tweak on this, or a tweak on that, that’s something that should be done to make the customer happy and satisfied. But if you drive your entire product strategy over what this or that customer needs, at the end of the day, you’ll end up going out of business because you won’t have anything that you can sell more than once.
MURRAY: Operations and finance teams at fabless companies sometimes have a somewhat adversarial relationship. How can a well-integrated ERP system help functional teams work together more effectively?
JONES: We see a lot of companies that want to put in a software system, but the feeling is “You’ve got to make the finance guys happy,” or “You’ve got to make the operations guys happy.” It seems there’s a natural supposition that the ERP system can’t make everyone happy. I don’t believe this to be true.
I believe that a well-integrated system is going to use all the same data, and it’s going to use it maybe in a little bit different form, but it’s all the same stuff. So, if there are yields, there are variances, and the finance people need to have that information. The engineer needs to know how many parts he or she lost, and what the bin-outs are. One shouldn’t have to satisfy one team at the expense the others.
Now, that being said, if you only put in one piece of software, and you put in an ERP package, then the finance guy is going to be happy because he’ll do his payables and receivables, but the operations guy is not going to be happy. The finance guy will say “Look, you’ve got a whole manufacturing module over here.” But the operations guy says “Yes, but I can’t use it.” So assuming that you put software in that is designed properly, there’s no need to make one side happy over another.
When problems arise, it’s usually because system selection is put under finance or operations, for instance, and they sort of selfishly say “Well, I’m going to choose a system that does what I need, and bag the other departments.” The solution to this is to choose a senior person who can really take the broad look. He or she drives the program, and he makes sure that the requirements that meet the most needs are addressed. Sometimes he will have to say “No.” This person needs to weigh all the requests, seeing the needs of the many or the biggest bang for the buck, and prioritize those needs over the one-off jobs that don’t make that much difference in the total scheme of things.
MURRAY: How long have you known the Tensoft team, and how did you first meet them?
JONES: I was first introduced to Tensoft in 2001. They had been doing Microsoft Dynamics GP implementations, not just for fabless companies, but for a variety of companies. So they were a top-flight G.P. implementer, but they really wanted to get into the fabless niche. They got the message, early-on, that the Microsoft Dynamics GP manufacturing module didn’t work for the fabless folks. With this knowledge, they said “OK. What does it take for us to be able to write some software that’s not just stand-alone, but can integrate with G.P. to deliver to customers something that satisfies their needs — not only in finance — but also in operations… out of the box?”
In the early days, the Tensoft folks’ experience was primarily in discrete manufacturing. We worked with them and advised them, and they caught on to the fabless industry’s needs at an amazingly fast rate. Over the years, they have built on that base to create their Fabless Semiconductor Manufacturing (FSM) package. It really is a very functional set of software that has been written from the ground up to implement with Microsoft Dynamics GP.
I believe the Tensoft folks have a grip on what it takes to do this. They’ve used this information to write the software, and they have the team in place to install the software in a timely fashion to make their customers happy.
MURRAY: Thanks, Ron, for an interesting conversation.
For further information, please contact Caprice Murray, Tensoft Director of Business Development, at 888-450-4030 x406 or at caprice@tensoft.com
