Intellectual Property (IP) Components

Table of Contents

I. Summary

II. Intellectual Property Components

A. IP Components - Definitions

B. Commercial IP Components

C. Introducing the IP Component Value Space

III. The User's Dilemma: Purchase or Produce IP Components

A. Build vs. Buy, Hard vs. Soft, Physical vs. Logical: The Tasks Involved

B. Build vs. Buy: The Decision Process

IV. Conclusion

A. IP Components - Definitions

The methodology change that has emerged to fix this time-to-market problem is design re-use. Instead of re-designing every part of every chip, re-use existing designs as much as possible and thus minimize the amount of new circuitry that must be created from scratch. The most prevalent and promising method of re-use is throughIP Components - pre-implemented, re-usable modules that can - in theory - be quickly inserted and verified to create a single-chip system. Variously called megacells, cores, Virtual Components, and other names, IP Components are a conceptual throw-back to breadboarding days when engineers could quickly stitch-together standardized chips to implement their system. The breadboard is now silicon; the tangible components are now electrical abstractions.

The shift towards IP Components has profound implications for the semiconductor industry. Most of today's component IC products (RAMs, graphics and networking ICs, etc.) will be replaced by IP Components that perform the same function on a single, extraordinarily complex IC. The magnitude of these shifts are staggering: for example, embedded on-chip memories will replace a significant portion of today's $50 billion memory IC market over the next decade. Instead of continuing to grow into a $200 billion memory chip business by 2005, the memory market will instead consist of $100+ billion in memory IP components and system-chip implementations, and under $100 billion in memory parts. The exact split is impossible to predict, but memory IP components alone will be a several billion dollar industry ten years hence.

Intellectual property is an object or intangible item - sound, picture, combination of bits - whose major value comes from the skill or artistry of its producer, and not from the tangible medium of its delivery, such as paper, CD, or bitstream. IP Components are, by definition, intellectual property: they are intangible, significant effort and skill/artistry goes into their creation, and the cost of the transmission medium is virtually nil. The effort and skill that goes into their creation is of two basic types: architectural / algorithmic design, and physical design and optimization. Different types of IP Components have more or less of each of these types of IP embedded within them. An industry-standard naming convention has emerged to differentiate the physical design content of IP Components (see also Table 1):

• Soft Components have the architecture and algorithm specified in an RTL (Register-Transfer Level) code description that can be read by a synthesis tool, but the design has virtually no physical information and has not been tuned to a semiconductor process;
• Firm Components are Soft Component that have been floorplanned (the basic design topology is established) and synthesized into one or more technologies to get performance, area, and power estimates;
• Hard Components are fully implemented into the mask-level data required to implement the block in silicon. The level of tuning to the process may be fairly low, as in the case of Generic IP Components, or exceptionally high, as in the case of process-specific components such as Artisan Components' Process-Perfect components.

B. Commercial IP Components

• Deeper because they need far more complicated components: Design teams used to combine relatively simple blocks such as UARTs, Z80s, and register files; now they require extremely complex blocks such as MPEGII, 486 cores, megabit cache RAM, and embedded DRAM, any one of which stand-alone would have been considered a complex chip.
• Broader because they need far more types of components: Semiconductor suppliers used to be able to specialize on particular subcomponent applications, and system suppliers would procure different components from different semiconductor "specialists." Thus, one semiconductor company would offer processors, while another would offer memory, a third would provide graphics chips, a fourth the communications chips, and so on. Today, however, the systems designer must be able to get all of these components from a single semiconductor supplier to integrate into a single-chip system. Since a single-chip system is almost always more cost-effective than a multi-chip system, a system designer is under strong pressure to integrate as much as possible, and the semiconductor supplier is under strong pressure to provide all of the required components.

A commercial industry has emerged seemingly overnight to supply the IP Component needs of the semiconductor industry. Although this growth caught many by surprise, it is in fact a natural step as the semiconductor industry - or any industry - matures. New industries start as completely self-sufficient and vertically integrated because their needs are specialized yet too small to interest suppliers. As the industry grows, it spins-off sub-industries around it. The semiconductor industry has seen many of these cycles, such as the emergence of the independent semiconductor equipment industry in the late-1970s and the independent EDA industry in the early-1980s. The commercial IP Component industry has emerged for all of the same reasons the previous two industries did:

• demand is large enough for an independent industry to emerge and be profitable;
• the requisite labor force is specialized, of finite size, and thus hard to find and retain;
• a focused supplier who can spread development costs over a broad customer base has a very strong competitive advantage over a captive supplier who can only spread costs across a narrow base;
• the value of multi-discipline expertise (e.g., knowing logic design and process and fabrication and testing) is of diminishing importance, because a Soft Components is incomplete, yet still can have high value (but implementation costs are still high);
• Standards are emerging which facilitate component interoperability and transfer.

An IP Component portfolio will contain many trivial or rarely used "filler" components that are necessary to be in the single-chip system business, but are not differentiators. Examples of this might be UARTs or Z80 blocks, where a superb implementation is of no higher value than a reasonably good implementation. This is in contrast to a much smaller group of components that are differentiatable and will bring the semiconductor supplier business he would otherwise not get. These differentiatable components - as a modern CISC processor, high-speed or low-power SRAMs, embedded DRAM, or an MPEG II core - are often the nucleus of the single-chip system, and it is extremely important that they be as fast/small/cool as possible. The specifications for these components will usually be the deciding factor when deciding among different semiconductor suppliers; in fact, the competitiveness and value of their entire component portfolio will often be judged first by the specifications of the "nucleus" component(s), and then whether all other required "filler" components are available.

C. Introducing the IP Component Value Space

For IP Components, though, "IP value" and "extent of legal protection" have virtually no correlation. IP Component suppliers are constrained by a labyrinth of century-old rules and restrictions about what is and is not patentable and copyrightable. For example, copyrights are easy to get for VHDL/Verilog source code, and on layout data. Patents, on the other hand, can be granted for innovative circuit designs, but not the layout. Furthermore, legal protection is granted on uniqueness only, with no concept or judgment made about the inherent worth of the patent or copyright. "Value" ultimately boils down to a simple statement: "How badly do you want this component?"

This is a fundamental attribute of IP in general, and IP Components are no exception: classical free-market economic forces of scarcity and desirability apply. Those laws state that the harder it is to find something (scarcity) and the more desirable it is, the more one will pay, independent of the costs of production. But how does one compare the value of a synthesizable MPEG core against a CISC processor physical core against a process-optimized RAM generator? Although the natural variables would be "desirability" and "scarcity," these attributes are impossible to measure. In particular, "desirability" tends to be very binary - either you need a component, or you do not. "Scarcity," on the other hand, can be easily represented by complexity, since the supply of a component drops as it becomes more complex. (Note that this is the only place legal protection might affect value: a legally enforced monopoly, also known as a patent or copyright, can make something very scarce.) "Complexity" has two dimensions to it, however: functional complexity, or how sophisticated is the algorithm and architecture, and physical complexity, or how optimally implemented in silicon (as measured by performance, size, and power) is the function. For example, a synthesizable RTL description of a Pentium core would have enormous functional complexity, but low physical complexity, while process-optimized RAM generators have a medium functional complexity but very high physical complexity. A fully routed processor core is high in both dimensions. This is the IP Component Value Map, as shown in Figure 1 on the following page. Value increases as one moves out in either the logical or physical complexity dimensions because one must hire expensive "artisans" for long periods of time to implement state-of-the-art components.

A few additional terms are introduced in Figure 1. Hard Components have been broken into two groups: Hard Cores and Physical IP Components, the latter including process-tuned on-chip memories, analog and datapath elements, standard cells, and I/Os. Another category called Generic IP Components is introduced. Generic IP Components are similar in type to Physical IP Components, but they have less process tuning. Physical and Generic IP Components are similar because they have greater physical complexity than functional complexity. Note, however, that Physical IP Components are exactly tuned to a semiconductor process, while Generic IP Components use a lowest-common-denominator physical and electrical rule set to make them portable to multiple processes. Examples of Physical IP Components are Artisan Components' Process-Perfect components, and libraries and compilers provided by semiconductor and ASIC manufacturers; examples of Generic Components are portable or automatically-sized libraries provided by commercial library suppliers such as Compass and Aspec. Physical IP Components are often the determinant of system performance and density. For example, the new 533 MHz Exponential Technology x704 PowerPC microprocessor has 2.23 million memory transistors, but only 0.47 million logic transistors. This is really a "Physical IP Component with a little logic sprinkled in."

Going across the IP Component Value Space are "isoquants," an economic term that means that any point on a single isoquant line has the same perceived value as any other point. Each isoquant up and to the right has higher value than lower isoquants. Thus, components that are high on one dimension but low on the other are of roughly comparable value, while components that are high in both dimensions are of significantly higher value. The Intellectual Property embedded in IP Components - and thus their perceived value - can be anywhere from slight to enormous. Just like packaged semiconductor products, an incredibly broad range of IP Components will become available that span the entire map, from simple to complex, from roughly coded to painstakingly laid-out, from cheap (or free) to extremely expensive.

For all other semiconductor companies, however, the needs are opposite: designs are usually fabricated at a single foundry (usually theirs), so they want their IP Components to be fully optimized for their process both physically and electrically. Optimization is crucial because even minor improvements can be significant: a 10% area reduction can be worth tens of millions of dollars in production costs over the life of a process. Therefore, they need Physical IP Components, not Generic IP Components. Furthermore, they support "customers" (external systems designers for ASIC suppliers; internal users for the integrated semiconductor manufacturer) whose IP Component needs are usually very broad and usually not predictable ahead of time. These semiconductor suppliers must create a broad, optimized IP Component portfolio ahead of demand so that the inevitable last-minute component additions ("You need a what?!? We don't have one!") do not overwhelm the organization.

Looking at the needs of the ASIC, merchant, and captive semiconductor suppliers, therefore, leads to the inevitable question: "Now that I see the imperative for an IP Component portfolio, what should I do about it?" The possible options are surprisingly limited: (1) do nothing, and hope the need goes away; (2) try to build them all; (3) try to buy them all; (4) try to find the optimal combination of building some and buying the rest. Semiconductor suppliers that choose option (1) must either focus on a specialized niche with limited IP Component needs, or they simply will go out of business. Option (2) is neither feasible nor practical, unless, again, a semiconductor supplier is focusing on a very narrow niche. Options (3) and (4) are the same theme, differing only in degree, and will be the preferred path for the vast majority of semiconductor suppliers.

Procuring IP Components from a commercial supplier is not a panacea, however. Even if a commercial supplier provides 100% of its IP Component needs, a semiconductor supplier must still expend enormous effort to make the components usable by its customers. The reason is simple: IP Components are not "shrink wrap" items. The problem is especially acute for Soft IP Components, where the algorithm and architecture are specified, but not implemented. As any IC designer knows, a great deal of work is required to transform RTL code into an acceptable placed-and-routed design. And, the challenges surrounding "deep submicron" design is making the implementation process proportionately harder. There is also the problem of user-level models: cycle-accurate behavioral, C language, and other simulation models must be created and functionally verified before the components is usable. Therefore, semiconductor suppliers will focus virtually all of their Central Library/ CAD efforts on transforming Soft components into Hard ones usable by its customers.

The reason is simple: most IP Component companies cannot create Hard components, both because they lack the expertise and it is not economical for them to implement the design for every customer. The majority of components will therefore be Soft, or possibly Firm. The semiconductor supplier has a choice: expend the resources to implement the component, or force its customers to do so when they design their system. But customers do not want to perform the implementation of the component because it eliminates most of the time-to-market value that IP Components offer. Far better for the system designers to find a semiconductor supplier that has already implemented the appropriate components. And they will - customers will naturally gravitate towards semiconductor suppliers that offer Hard components. Semiconductor suppliers without a full portfolio of Hard components will find themselves at an enormous competitive disadvantage.

A strong Central Library /Central CAD organization is still required to make IP Components usable by designers; indeed, even if all components come from commercial suppliers, most Central Library groups will see their workload increase when productizing all of the necessary components. Semiconductor suppliers therefore have a fundamental decision to make: since they cannot do it all, what should they build internally, and what should they buy? And how much effort does it really take to productize an IP Component?

A. Build vs. Buy, Hard vs. Soft, Physical vs. Logical: The Tasks Involved

The tasks are significantly different between Hard and Soft Components, as shown in Table 2. However, there is an even bigger difference when productizing Physical IP Components such as memory generators, cell libraries, I/Os, datapath blocks, analog blocks, and other "not synchronous digital logic," as shown in Table 3. Clearly, the procurement decision that offers the least work and the best time-to-market for the Central Library group are commercially-supplied Physical IP Components; the most work and the longest time-to-market are for the group to do everything internally, or to purchase Soft components and then implement them.
 

Table 2: Design tasks required of a Central Library organization to productize a synthesizable IP Component, depending on whether the component is built internally or procured in "Soft" or "Hard" form. 

Semiconductor Supplier's Tasks to Productize SYNTHESIZABLE IP Components	Tasks When Building Internally	Tasks When Purchasing Soft Core	Tasks When Purchasing Hard Core
oversee, negotiate with suppliers	-	X	X
design architecture	X	-	-
verify algorithm	X	-	-
write, test RTL code	X	-	-
design, test block interfaces	X	-	-
synthesis, floorplanning, place-and-route	X	X	-
extraction, timing modeling and simulation	X	X	maybe
design rule checking (DRC)	X	X	maybe
high-level model creation	X	maybe	maybe
design view generation - VHDL, Verilog, Synopsys, timing models, others	X	only timing models	maybe
datasheet generation	X	maybe	maybe
test pattern generation	X	maybe	maybe
integration testing	X	X	X
test instance fabrication	X	X	X

 

Table 3: Design tasks required of a Central Library organization to productize a PHYSICAL IP Component (memories, standard cells, I/Os, etc.), depending on whether the component is procured commercially, or built internally. 

Semiconductor Supplier's Tasks to Productize Physical IP Components	Tasks When Building Internally	Tasks When Purchasing From Commercial Supplier
oversee, negotiate with suppliers	-	X
design architecture (memories)	X	-
design layout (cell height, grids, etc.)	X	-
build generator infrastructure for layout (memories)	X	-
build generator infrastructure for design views and sim. models (memories)	X	-
optimize circuits	X	-
optimize layout	X	-
extraction, timing simulation	X	-
design rule checking (DRC)	X	-
timing modeling	X	-
design view generation - VHDL, Verilog, Synopsys, timing models, others	X	-
datasheet generation	X	-
test pattern generation	X	-
integration testing	X	minimal
test instance fabrication	X	X

B. Build vs. Buy: The Decision Process

Physical IP Components such as Artisan Components' Process-Perfect memory generators and standard cells should be the first type of commercial component procured. Nearly all of the development work is done by the component supplier, so that a high-quality Physical IP Component can be in designers' hands faster, cheaper, and far more easily than if the semiconductor supplier developed it in-house. Soft components are a "necessary evil," since most IP Components will only be available in this form. The daunting work of implementing, testing, and productizing these components will transform semiconductor companies' Internal CAD/Library groups into a combination of "Soft Core implementers" and "IP portfolio managers"; they will not have the time, skills, nor inclination to build their own Physical IP Components.
 

Last modified: Mon May 5 11:52:14 PDT 1997 
Copyright © 1997 Artisan Components, Inc. All rights reserved.