January 16, 2006
Kilopass Finds a Niche in Embedded Non-Volatile Memory (NVM)Technology
Please note that contributed articles, blog entries, and comments posted on EDACafe.com are the views and opinion of the author and do not necessarily represent the views and opinions of the management and staff of Internet Business Systems and its subsidiary web-sites.
| by Jack Horgan - Contributing Editor
Posted anew every four weeks or so, the EDA WEEKLY delivers to its readers information concerning the latest happenings in the EDA industry, covering vendors, products, finances and new developments. Frequently, feature articles on selected public or private EDA companies are presented. Brought to you by EDACafe.com. If we miss a story or subject that you feel deserves to be included, or you just want to suggest a future topic, please contact us! Questions? Feedback? Click here. Thank you!
In early October Kilopass Technology Inc. completed its second round of venture capital funding for $8.8 million. The firm was founded in 2001 with the mission to become the industry leader in embedded non-volatile memory technology (NVM) a niche within memory market. According to a report by BCC (Business Communications Company) NVM market is estimated to be $17.4 billion in 2005 with a projected average annual growth rate of 32% though 2010. Flash memory accounts for almost 90% of the total NVM market. EPROM (Erasable Programmable Read-Only Memory) and EEPROM (Electrically Erasable Programmable Read-Only Memory) negative growth. The emergence of advance memory technologies such as
MRAM, FRAM and OUM are expected to gradually take away some share of total volatile memory.
I had an opportunity to discuss Kilopass with Charles Ng, Vice-President of Worldwide Sales & Marketing. Mr. Ng has more than 22 years of experience in the semiconductor industry. Prior to joining Kilopass, he held senior management and executive-level positions with Cadence, Compass Design Automation, and VLSI Technology Inc. He was the key contributor in establishing and expanding the business for these companies in North America, Japan, and Asia Pacific. He graduated from California Institute of Technology with a MS in Computer Science and a BS in Computer Engineering and an MS in Computer Science, and he also holds a MBA from UC at Berkeley.
Were you a founder or an early employee at Kilopass?
One of the very early employees.
Why did you leave Cadence, an industry leader, to work for this startup?
Opportunity! Good technology. Good startup.
What business is Kilopass in?
Kilopass is an IP Supplier of field programmable memory. You can program the memory content. It is an IP embedded in a SoC. Non-volatile memory which means that data content is kept after the power is off. Our claim to fame is that we require only standard logic CMOS process without any additional processing at all. Volatile memory requires a complicated process which is very expensive to develop and to use.
We can avoid that cost using standard CMOS process technology. Our technology is very cost effective because of standard CMOS logic. It's available everywhere and can be found in many foundries. Our technology supports 90 nm and 65 nm in the future much easier than flash memory. With flash memory it is much more difficult. They do not scale very well into advanced processes.
We can be very high density. We are shipping up to 1 Mbit. Other solutions that can provide similar capabilities can only provide about 1 Kbits or so. Our solution can go up to 1 Mbit, while this may not seem very high, it is high relatively speaking.
It is very secure which means that it is very difficult for anyone to break into the memory, to figure out the memory content. It prevents reverse engineering and protects memory content.
Editor: The flash types of non-volatile memory are limited in their ability to store charge due to the fact that as the logic oxides get thinner, direct tunneling occurs and the charge tunnels off. This occurs between 80 and 85 Angstrom for the thickness of the tunnel oxide required to hold the charge. Logic CMOS gate oxides are much thinner, currently in the range of 30 Angstroms. Moreover, the lower voltage required to achieve tunneling limits the pacing between storage elements (transistors).
Tell me a little about Kilopass Technology.
We started in 2001. We are headquartered in Silicon Valley. We are funded by US Venture Partners (USVP) our lead VC followed by BuleRun Ventures who used to be called Nokia Venture Partners and by InveStar Captial who is related to TSMC (Taiwan Semiconductor Manufacturing Company) and iGlobe Partners. Our current headcount is around 35 people in the US. We also have an outsourced location and they have about 35 people also. So altogether we have about 70 people. We have 14 patents granted with many more pending.
Where is that outsourced location?
That's in mainland China!
What is the Kilopass' annual revenue?
A few million dollars!
How many customers does Kilopass have?
About 30 customers!
What process nodes do you support?
The memory module is currently available in 0.18-micron, 0.15-micron and 0.13-micron. We have working silicon and passed 1,000 hours of burn-in testing. This is a very rigorous test for our IP. We just had a press release with TSMC announcing that they have certified XPM on their 0.18-micron CMOS logic process as part of their certification program. We are very proud that we can achieve that. We are shipping to customers at process nodes 0.18-micron to 0.13-micron.
What are the key features of the technology?
It is a new type of field programmable non-volatile memory. We are the only using standard CMOS, which is very low cost. It is very scalable. It can be high density up to 64 megabits. Theoretically it can go higher to .5 Gbits or so. But today we are shipping up to 1Mbit. We can offer you tens of megabits. It is not a limitation of the technology. It is more a limitation of engineering resources. If we were to put enough engineering resources into it, we could develop much larger memory blocks. It's a very secure memory that prevents anyone from trying to reverse engineer. The read voltage is very standard and programming and standby currents are very low. Very good reliability
and data retention. That's very important in non-volatile memory.
What does XPM stand for?
eXtra Permanent Memory
High density layout. Each memory cell takes up less than 1½ T, minimum size transistor. It has very low programmable current which makes it ideal for field programming; 100 ìA/bit programming current and read current around 15 ìA/bit. You can program anywhere: on the production floor, inside a package, or inside a printed circuit board system. It is one time programmable on the device level but actually it is a limited time programmable solution on a system level due to multiple sectors. If you need only 1,000 bits of memory, you can put in slightly larger memory, say 8,000 bits. With 8,000 bits, you now have 8 different sectors. You can program 8 times the memory
content of 1,000 bits. Because the die size difference between 1,000 bits and 8,000 bits is very small, the penalty is very small. With this kind of approach, you can have a limited time, a few times programmability.
There is very good data retention - over 20 years. It is very reliable for mission critical applications. Very secure from any reverse engineering.
What are some of main applications for XPM?
One application is firmware storage. Anytime you have an embedded MCU (Microcontroller Unit) like an AD51 or ARM processor, you need to store the firmware somewhere. If the firmware needs to be programmable, if you need to change it often or if you have many different versions, then you want to have the flexibility to program it in different ways. Today people usually put that firmware into a separate chip, an off chip flash chip that is programmable. You would download that firmware during boot up time into the SRAM block inside your SoC chip and then the MCU would access the memory content from the SRAM. But this two chip solution can be transformed into a single chip solution with
XPM. This is because XPM is non-volatile, it is field programmable and it only requires CMOS so that you can easily integrate it into you SoC chip. With this approach you eliminate the external flash chip with a single chip solution. XPM is much smaller in die size than SRAM. Usually a smaller amount of SRAM is needed but that depends on the overall design. So the whole die size shrinks down a lot. It provides instant boot up because there is no more downloading of the firmware during boot up. It is very secure storage because now the firmware is inside the SoC chip. It's not outside. Even when people try reverse engineering, peeling the layers off one layer by one layer, it is
still very difficult to detect the memory content using a microscope or thermal emission equipment. Since it is a single chip solution, it reduces the overall form factor. That's one application.
Another application is use in smart card MCU. The MCU needs to store firmware. Also some smart cards often need to store security information like personal identification data, biometric data, health data and all that.
It can also be used for storing low density content. This is usually less than 1,000 bits. There is a growing trend for SoCs to store id number somewhere in the chip. Our solution is perfect for that because it is non-volatile and uses standard CMOS. There are many applications like digital content protection to protect songs, movies and all that. Under digital content protection there is a standard called HDMI (High Definition Media Interface). This also needs to store security code as well.
Editor: HDMI is the first and only industry-supported, uncompressed, all-digital audio/video interface. HDMI provides an interface between any audio/video source, such as a set-top box, DVD player, or A/V receiver and an audio and/or video monitor, such as a digital television (DTV), over a single cable. HDCP (High-Bandwidth Digital Content Protection) is a content protection technology available for use in connection with HDMI that was developed by Intel Corporation with help of Silicon Image.
Another possible application is analog trimming. Any analog chip can use field trimming and field calibration. With this flexibility it can really improve the performance of analog chips a lot. Our solution is ideal for that.
Would you expand upon that?
The problem with analog chips is that the performance can be very low due to process variations. The analog chip can not perform optimally because it doesn't know when it comes out of silicon where it is in the process corner. You can test the silicon in the field and memorize some of the calibration data you get, in a few bits. You can look at the silicon and find out where it is in the corners that is fast corner or slow corner. You can do self-calibration and self-trimming based upon the process corner and improve the performance of the analog chip.
Editor: Traditional analog trimming techniques include laser trimming of thin-film resistors, laser cutting of metal or polysilicon links, zener-diode zapping to adjust resistance, opening polysilicon fuses with a current, and storing bits in an EPROM or EEPROM to control a DAC to adjust currents or voltages. Each of these techniques has its own shortcomings.
LED display driver IP basically has an analog component. This is very similar to that is CMOS sensor which is an analog IP for digital cameras and cellular phones that requires field calibration capability. We are ideal for that solution.
You can also use XPM for configuration. You can set a few bits to self-configure the chip to turn on or off certain features. You can use it to store a unique chip id code to track production by part number. You can use it for memory repair to mark defective rows or columns in embedded SRAM to improve SRAM yield. Low density application is very broad, different kinds of ICs.
Would you summarize?
It is a new type of non-volatile memory based upon standard CMOS logic process.
Very scalable. It can go down to 90 nm and beyond. We will be working on 65 nm in the near future. It can be a very low cost solution. It can be very secure. Memory content is well protected. Low density for ids and high density for firmware.
What is the primary marketing target for the product?
It is very broad. In low density memory storing security codes and keys is a growing trend. Many different applications want to store codes e.g. content protection for digital entertainment. Analog IP is very broad - almost any kind of analog would have process variation issues and need field trimming and calibration. Firmware storage - embedded MCU is everywhere, used for many different applications.
In addition to flash memory is there any other type of alternative approach out there in the market?
Embedded flash is not standard logic design. There is a big difference in cost, density and scalability. Embedded flash is a much more difficult process. It is still on top of CMOS but a lot more expensive and a lot more difficult to develop and because of that not available in many places whereas standard logic CMOS is available in many different foundries because it is the most commonly used technology.
Do you see mask programmable ROM as an alternate?
By definition it is not field programmable. You need the mask set. Changing the mask set is much more difficult to manage. It makes it a totally different application. Our solution is field programmable. You can program it any where. That drives the application into a total different space.
Is there any competitor with similar technology?
Actually, if you look at CMOS embedded non-volatile memory and only CMOS with scalability, security and high density, then we are really the only solution provider that can offer all of those features.
Is there anybody else with many of those features?
Nobody can offer that combination. There are some companies like Virage that offer a solution.
The company hired Bernie Aronson as CEO last fall. What did he bring to the firm?
Bernie Aronson joined about a year ago. He is a very successful entrepreneur. He brought two companies to their IPOs. He was president of EPIC Design Technology many years ago and brought them to an IPO. He also joined Synplicity and brought them to an IPO. He was also founder and president of Pico Design. We are very fortunate and very happy to be able to recruit him to join us.
Has he changed the direction of the company?
After he joined us, we had the second round of venture funding where US Venture Partners came in. We are continuing with the product development and continuing to expand the market.
How do you sell your product? Direct sales, VARs, distributors?
We have both a direct sales force and distributors around the world.
Is that country dependent?
Yes. We are mainly selling direct in the US. We have distribution in Asia, Japan and Europe.
How do you price your product?
Anywhere from $100K. Some contracts can be multiple millions of dollars. It depends upon the business arrangement.
Is there a licensing fee and a royalty component?
Yes, there are both. The license fee depends upon the memory density.
You said that the technology could support a much higher capacity than your current offering. Is the greatest market opportunity at relatively low memory density?
Actually, at both ends. There is a lot of demand in low density and there is also demand in high density. I would say that is all across the map. So when we ship from 16 bits to a few megabits, it covers the whole market really well.
Are your sales concentrated in any one geography?
Today it is more concentrated in the US.
Do you expect that to change over time?
Yes, of course! It is more difficult to develop an international market. Among our customers is Accent, one of the largest design service company in Europe, and Global Unichip in Taiwan.
Editor: Accent, founded in 1993, is a joint venture between STMicroelectronics and Cadence. Accent is based in Europe, with offices and design facilities in Italy - Milan and Genoa - and in France. Accent staff includes more than 120 engineers. Accent is a leading design house currently ranked 6th in Europe by market research firm Future Horizons. Global Unichip Corporation is a full service SoC Design Foundry, was founded in 1998 at Hsinchu Science Park, Taiwan. In 2003, TSMC became its primary shareholder.
How do you deliver and protect your IP?
We deliver it in electronic form, an electronic data base. We protect it by a
legal contract. There is no physical protection of the IP. There's a packing slip. You know the IP industry. We are just like any other IP vendor. They put a tag on the IP. We do the same thing.
Legal protection in some countries is not as great as in the United Sates. Does that give you any concern?
Interesting question! I just came back from a conference on IP,
IP/SOC 2005 (IP Based SoC Design) in Grenoble. Almost all of the people in the industry agree that if you think about, it is not a problem. The legal protection is adequate for most of our business. And I totally agree. Legal protection is good enough and it addresses a big chunk of market already. I don't need to know about the tiny percentage of the market that is not protected. All my markets including Asia are protected. There are some countries in Asia where it is not well protected. But the majority of the market in Asia that is of interest to me is legally protected.
What is the size of your target market, the total available market (TAM)?
Hundreds of millions of dollars!
If you are now at a few million dollars without any significant competition, what challenges does Kilopass have to overcome to get to that level?
Execute carefully to expand engineering to develop more products and to expand sales and marketing. I see it that we just need to execute well. I don't see any fundamental barriers to our growth.
You are confident that the technology will scale to new process nodes and not be obsolesced?
Yes. I see the challenge as mainly managing resources - business and engineering.
How do you convince a prospect about XPM - demo, benchmark, reference account, ..?
It is a normal sales process. We explain to them how it works. It is not any different from selling other high technology products. You always need to convince a customer that this is what it does. The need is there. We don't have to convince prospects.
How long would a prospect take to evaluate your technology?
The sale cycle is a few months. We are usually successful if our technology is a good fit.
Letters to the Editor
From Zuken re article on Mentor Graphics Expedition Enterprise Flow
With regard to the statement by Mr. Weins, "Our capability with design data management is much more significant that what either Zuken or Cadence has provided. Our ability to tie into corporate enterprise systems is more significant.", it is worth noting that Zuken was the first EDA company to enter the PLM market in 1995 and has more than 100 customers using our enterprise-level EDA solution coupled directly with our own PLM solutions. Zuken has partnerships with companies such as SAP to provide seamless integrations with ERP and has numerous customers in production with such enterprise-level solutions - sharing data and work flows among engineering, manufacturing, and purchasing. We
have a solid story backed by over a decade of success.
From Barney Biancavilla of NGC
A lot of our board's sizes and shapes are custom from a 3D mechanical system
IDF is wholly inadequate data design transfer.
Mentor's Expedition Enterprise Flow is completely self serving AGAIN and not real world.
Completed Boards must be re-evaluated back in 3D design.
The top articles over the last two weeks as determined by the number of readers were
European Project to Address Power Leakage in Digital SoCs
A consortium of 14 partners, universities and research institutes from seven European countries has launched a project to enhance the design of the next generation of SoCs by addressing the key issues of leakage power in CMOS process of 65nm and below. Named CLEAN (Controlling Leakage power in NanoCMOS SoCs), the project is sponsored by the European Commission with a grant of 4.5 million euro, under the "Nanoelectronics" Strategic Objective of IST 6th Framework Programme.
SMIC Adopts Mentor Graphics Eldo Simulator for Analog Circuits for its 0.13-micron and Below Process Nodes
Semiconductor Manufacturing International Corporation has adopted Mentor's Eldo simulation tool as an internal SPICE simulator for analog circuits. In addition to 0.18-micron, SMIC will now provide Eldo device models on SMIC's 0.13-micron and smaller process technologies aimed at satisfying the design needs of Mentor's and SMIC's mutual customers.
, makes course materials used in the teaching of all MIT undergraduate and graduate subjects available on the web, free of charge, to any user in the world.
Simucad Spins Off From Silvaco With Sights on IPO
Simucad will focus on delivering its complete line of PDK-enabled analog, mixed-signal and RF IC design software. Simucad's corporate objective is to build its revenue stream required for an IPO as soon as possible.
Azuro Secures $9M in Second Round Funding
140 Azuro was founded in 2002, and has raised a total of $13.3M in venture funding since its formation. The company's flagship product, PowerCentric(TM), is a complete solution for the implementation of low power clock networks on digital semiconductor chips.
Other EDA News
Other IP & SoC News
You can find the full EDACafe event calendar here
To read more news, click here
-- Jack Horgan, EDACafe.com Contributing Editor.