Reconfigurable Systems Emerge

Reconfigurable Systems Emerge
The Rise of Reconfigurable Systems 27 March 2017 Reconfigurable Systems Emerge Reconfigurable Systems Emerge. Nick Tredennick, Editor Gilder Technology Report 27 March 2017 Nick Tredennick Nick Tredennick

The Rise of Reconfigurable Systems
27 March 2017 Overview Major trends affecting the microprocessor market Value PC Value transistor Emerging economies Microprocessors Computer microprocessors Embedded microprocessors Configurable microprocessors PLD microprocessors Three major trends affect the microprocessor: the value PC, the value transistor, and emerging economies. I’ll divide microprocessors into four categories so I can say, by manufacturer, what will happen in each area. The areas are: computer microprocessors, embedded microprocessors, configurable microprocessors, and PLD microprocessors. Computer microprocessors are microprocessors used in computers. Embedded microprocessors are essentially invisible in their systems. That is, the user doesn’t need to be aware that there’s a microprocessor in the system. Configurable microprocessors have instruction sets that can be altered to orient them to a particular application. PLD microprocessors are implemented inside programmable logic devices either as a hard core or as a soft core. These categories are not necessarily exclusive, but they are convenient for analysis. 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 Moore’s Law 27 March 2017 Nick Tredennick Nick Tredennick

Top View: Field-Effect Transistor
The Rise of Reconfigurable Systems Top View: Field-Effect Transistor 27 March 2017 This figure is more background for conclusions I’ll draw later. It illustrates where we’ve been in semiconductor processing and where we’re headed. I wanted to remind you of the dimensions semiconductor companies work with. Looking at the 1991 transistor on the left, it is made up of a “channel” and a “gate.” Current flows through the channel. The gate, which sits on top of the channel, controls the current flow. The width, or “line width,” of the channel is 750 nm, the line width of the gate is also 750 nm. Because making transistors is essentially a two-dimensional process, cutting the line width in half enables four times as many transistors in the same area. Four of the 370-nm transistors fit in the area occupied by the 750-nm transistor. By 1991, transistors were already smaller than bacteria. By the 370-nm generation, transistors were smaller than the wavelength of ordinary light. You cannot see these transistors with an optical microscope. This year, companies will build chips with 90-nm transistors. That’s twice the size of a virus. At 90 nm, 100,000 transistors fit on a small grain of sand! That’s why big chips have hundreds of millions of transistors. 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 The Microprocessor 10 years of Moore’s-law progress led to the microprocessor The second generic component Raised engineers’ productivity Problem-solving became programming Grew to billions of units/year Stalled progress in design methods for thirty years 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 The Personal Computer 10 years of microprocessor progress led to the PC Dominated the industry for 20 years Supply of performance grows with Moore’s law Demand grows more slowly Diverging growth in supply and demand leads to the value PC 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 The PC Is Good Enough The PC is good enough. For a long time, the PC didn’t enough performance to satisfy anyone. Industry inertia is on performance and assumes that function comes in the wake of delivering performance. The big change is that for the first time we can see a point when there’s enough performance for the vast majority of users. The quest for performance will no longer drive development. 27 March 2017 Nick Tredennick Nick Tredennick

The Path To The Value Transistor
The Rise of Reconfigurable Systems 27 March 2017 The Path To The Value Transistor I’m using the transistor to illustrate supply and demand. The concept of supply and demand is underappreciated in the semiconductor business because the industry has been growing at the leading edge of what Moore’s law could supply that we have come to take it for granted that supply and demand are synonymous for our industry. It isn’t so. When the transistor came out, it wasn’t good enough for any of its applications. The transistor improved at some rate. The demand for transistor performance grew also, but at a different rate. There’s no necessary correlation between the rate of improvement in supply and the rate of increase in demand for what is being supplied. These phony curves illustrate the point. The transistor began with performance well below demand. Over time, the transistor improved. Demand for performance rose, but it also spread—and it increased at its own rate that was independent of the rate of improvement in transistors. After a few years, the transistor had improved enough and demand had spread enough that there were some transistors that were good enough for some applications. The same arguments apply to integrated circuits. There’s a huge difference in the performance demanded by leading-edge digital transceivers and by consumer washing machines, for example. Washing machines don’t need leading-edge ASICs. 27 March 2017 Nick Tredennick Nick Tredennick

Transistors Are Good Enough
The Rise of Reconfigurable Systems Transistors Are Good Enough 27 March 2017 Chip power depends on the total number of transistors, on how many transistors are “active,” on the semiconductor process, and on the chip’s operating frequency. What this model shows is that, for this example application, the best transistor for the job, something I call the “value transistor,” is between 80 and 90 nm, depending on frequency. What’s new is the emergence of the “value transistor”—that there even is one—not its particulars. In the ’80s, for example, no one imagined that the PC would ever reach users’ performance requirements, but today we see that the value PC satisfies the needs of most users. The same will be so for the value transistor. It’s not its absolute size or even whether it has one, only that for an increasing number of applications, there is a transistor that is the right one. Big transistors use too much active power and small transistors leak too much, so there’s a transistor in the middle that’s right for this application. Follow the 100-MHz line. It’s second from the bottom. This graph shows a 200-million-transistor chip that dissipates 15 watts in a 130-nm process. At 130 nm, the active power of the chip’s big transistors dominates total power. As the transistors shrink, power decreases to about 6 watts in a 90-nm process. As the transistors get smaller, active power continues to decrease, but the smaller transistors leak more and soon leakage power dominates. Total power rises to 12 watts or so. This chip’s value transistor is in a 90-nm process. For a long time, transistors were too big for most applications. Everyone helped pay for improvements. With the emergence of the value transistor, some applications have the transistor they need. The value transistor for a particular application might occur at 90 nm, 130 nm, 180 nm, or, as we shall see, even at 250 nm. Applications that have found their value transistor won’t help pay for process improvements. On the left side of this graph, active power dominates the chip; on the right side of the graph, leakage power dominates. As the number of transistors per chip increases and as transistors get smaller, more applications will have a “best transistor” that is not the smallest available transistor. 27 March 2017 Nick Tredennick Nick Tredennick

Foundries: Adoption Rate By Process
The Rise of Reconfigurable Systems 27 March 2017 Here’s evidence that some applications are finding their value transistor. This figure is a computer model I made to mimic a chart from TSMC’s web site. I built a model because I don’t own TSMC’s chart and because TSMC’s chart has a nonlinear scale. This chart shows the percent of the foundry’s wafers, called “wafer starts,” by semiconductor process, plotted against time. Production capacity is measured by the number of wafers a plant processes per month. To read the chart, draw a vertical line through a year. Draw horizontal lines where the process wedges cross that line to read the percent of wafer starts. For example, draw a vertical line at the year The 180-nm process was about 10% of wafer starts, 250 nm was a little over 20%, and 350 nm was about 35%. The plot has two characteristics that caught my attention: semiconductor-process adoption-rates are falling and old processes don’t phase out with time. Before the 1996 introduction of the 350-nm process, 100% of the foundry’s wafers were at 500 nm or larger. By the beginning of 1997, more than 30% of wafer starts were at 350 nm. In 1997, the foundry offered a 250-nm process, but by the end of the year, fewer than 20% of its wafer starts were at 250 nm. The foundry offered a 180-nm process in It grew to only about 5% of wafer starts by the end of the year. This is an interesting trend: with each new generation of semiconductor process, the adoption rate is falling. Look at the wedge for 250 nm. Even though the process was introduced in 1997, and even though three new processes have been introduced since then, the 250-nm process is still holding its own as a percentage of wafer starts. Since wafer starts are growing, demand for the 250-nm process is growing with the market. These curves represent demand and demand forecast for foundry processes. At an integrated device manufacturer, or IDM, such as AMD, Intel, or Texas Instruments, process adoption is set by corporate fiat and is an integral part of the company’s business model. Modeled after: TSMC 27 March 2017 Nick Tredennick Nick Tredennick

Semiconductors: Industry In Transition
The Rise of Reconfigurable Systems 27 March 2017 Semiconductors: Industry In Transition Causes The transistor is good enough The PC is good enough Effects Shift from tethered to mobile systems Changes design emphasis From: cost-performance To: cost-performance-per-watt Non-volatile memory will emerge Wafer stacking will emerge MEMS will emerge The semiconductor industry is due for a massive transformation. Non-volatile memory is on the way, wafer stacking is on the way, the personal computer is good enough for most users, and transistors are good enough for most applications. Industry emphasis is shifting from tethered to untethered systems, bringing with it a shift in the system design goal from cost-performance to cost-performance-per-watt. The microprocessor isn’t efficient enough to be the workhorse in untethered systems. The microprocessor will move to a supervisory role, its role as workhorse usurped by more direct alternatives. Similarly, the discrete components in today’s untethered devices aren’t good enough and will be displaced by microelectromechanical systems (MEMS). The transformation, which requires doing something new and different, will be slow and painful because the semiconductor industry is busy doing what it already knows how to do. The move from tethered to untethered systems will precipitate an equally disruptive transition in sensors, actuators, and discrete components (switches, capacitors, inductors…). The sensors, actuators, and discrete components that dominate the market today will give way to microelectromechanical systems (MEMS). MEMS are what they sound like—combinations of mechanical and electrical elements where the mechanical elements are scaled to the size of the electronic elements. MEMS can be made with the same equipment used to make semiconductor chips. The MEMS market, which was about $5 billion in 2002, is expected to double by For comparison, the semiconductor market was about $141 billion in 2002 and is expected to grow to $350 billion by 2007. Further, MEMS match in scale the associated electronic systems that make use of them. Their small scale makes them more sensitive, more efficient, cheaper, lighter, faster, and more durable than macro-scale equivalents in sensor and actuator applications. MEMS will be better for mobile applications than their macro-scale equivalents. The MEMS industry will grow even faster than the semiconductor industry that preceded it. 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 Design Alternatives What Value Who Microprocessors $40B Programmers ASICs $30B Logic designers FPGAs $3B Logic designers Here’s the market situation for FPGA vendors. The programmable logic market is about $3 billion. FPGAs and microprocessors are eating into a declining ASIC market. Microprocessors and their derivatives will win. FPGAs and microprocessors are usurping a declining ASIC market. Microprocessors (and their derivatives) will win. 27 March 2017 Nick Tredennick Nick Tredennick

Programmers And Logic Designers
The Rise of Reconfigurable Systems 27 March 2017 Programmers And Logic Designers Programmers optimize software Languages OS Compilers Applications The Users Manual is the (problematic) bridge Programmers Logic designers optimize hardware Microprocessors Memory Logic designers 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 Microprocessors Microprocessor advantages Flexibility High-volume production Usable by programmers Microprocessor limitations Too slow Too much power 27 March 2017 Nick Tredennick Nick Tredennick

Microprocessors Are Unsuitable
The Rise of Reconfigurable Systems 27 March 2017 Microprocessors Are Unsuitable 27 March 2017 Nick Tredennick Nick Tredennick

Application-Specific Integrated Circuits
The Rise of Reconfigurable Systems 27 March 2017 Application-Specific Integrated Circuits ASIC advantages Best performance Smallest chip The benchmark for function efficiency ASIC limitations Inflexible Expensive to design High fixed costs require large production runs Requires logic design 27 March 2017 Nick Tredennick Nick Tredennick

Programmable Logic Devices
The Rise of Reconfigurable Systems 27 March 2017 Programmable Logic Devices PLD advantages Flexibility High-volume production PLD limitations Chips too expensive Too slow Requires logic design 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 ASICs and PLDs (FPGAs) ASICs and PLDs are competing in a $30-billion market This competition will not cross into the microprocessor market because designs require logic designers 27 March 2017 Nick Tredennick Nick Tredennick

Supply and Demand: ASICs & PLDs
The Rise of Reconfigurable Systems 27 March 2017 Supply and Demand: ASICs & PLDs 27 March 2017 Nick Tredennick Nick Tredennick

Microprocessors and ASICs
The Rise of Reconfigurable Systems 27 March 2017 Microprocessors and ASICs Microprocessor For the ultimate in flexibility, programmers map the application onto a general-purpose microprocessor. We need a little background for the next microprocessor segment. Microprocessors offer the ultimate in flexibility and in high-volume production. Microprocessors allow programmers to map applications into working systems. Application-specific integrated circuits offer the ultimate in performance. Logic designers map applications into a custom circuit. Programmers Application For the ultimate in performance, logic designers map the application into a custom circuit. ASIC Logic designers 27 March 2017 Nick Tredennick Nick Tredennick

Microprocessor Evolution
The Rise of Reconfigurable Systems 27 March 2017 Microprocessor Evolution Microprocessor Design-time configurable microprocessor Run-time reconfigurable microprocessor Dynamically reconfigurable microprocessor ASIC FPGA ARC MIPS Tensilica Stretch Ascenium The problem-solving method available to programmers is programming. The problem-solving method available to logic designers is logic design. They do not mix. ASICs and FPGAs inhabit the world of the logic designers. Microprocessors inhabit the world of programmers. This chart shows the genealogy of microprocessors from ARC, MIPS, Tensilica, Stretch, and Ascenium. Notice that design-time configurable microprocessors are related to ASICs in that they require logic designers. Run-time configurable microprocessors are related to programmable logic devices (field-programmable gate arrays). A huge advantage goes to run-time configurable microprocessors and to dynamically reconfigurable microprocessors because they do not require logic designers and because they are generic at manufacture and are customized in the field. Programmers Logic designers 27 March 2017 Nick Tredennick Nick Tredennick

Design-Time Configurable Microprocessor
The Rise of Reconfigurable Systems 27 March 2017 Design-Time Configurable Microprocessor Most of the application runs as execution of general-purpose instructions Profile the application Programmers Application Create custom hardware and instructions to accelerate critical application sections Design-time configurable microprocessor Logic designers 27 March 2017 Nick Tredennick Nick Tredennick

Design-Time Configurable Microprocessor
The Rise of Reconfigurable Systems 27 March 2017 Design-Time Configurable Microprocessor Profile the application Create custom instructions for critical code sections Build specialized execution units Can be 10 to100 times faster than a general-purpose microprocessor on the target algorithm Examples: ARC and Tensilica Customized microprocessor limitations Requires logic designers Creates an application-specific, limited-function microprocessor Accelerates only critical sections 27 March 2017 Nick Tredennick Nick Tredennick

Run-Time Reconfigurable Microprocessor
The Rise of Reconfigurable Systems 27 March 2017 Run-Time Reconfigurable Microprocessor Most of the application runs as execution of general-purpose instructions Profile the application Run-time reconfigurable microprocessor Create custom instructions in an FPGA fabric to accelerate critical application sections Application Programmers Logic designers 27 March 2017 Nick Tredennick Nick Tredennick

Run-Time Reconfigurable Microprocessor
The Rise of Reconfigurable Systems 27 March 2017 Run-Time Reconfigurable Microprocessor Build a general-purpose microprocessor with integrated FPGA fabric Profile the application Create custom instructions for critical code sections Build custom execution units in FPGA fabric Can be 10 to 100 times faster than a general-purpose microprocessor on the target application Example: Stretch Run-time reconfigurable microprocessor limitations Accelerates only statically identifiable critical sections Limited to problems for which profiling works Profiling is difficult 27 March 2017 Nick Tredennick Nick Tredennick

Dynamically Reconfigurable Microprocessor
The Rise of Reconfigurable Systems 27 March 2017 Dynamically Reconfigurable Microprocessor Dynamically reconfigurable microprocessor Create custom instructions in a custom fabric to accelerate the entire application Application Programmers Logic designers 27 March 2017 Nick Tredennick Nick Tredennick

Dynamically Reconfigurable Microprocessor
The Rise of Reconfigurable Systems 27 March 2017 Dynamically Reconfigurable Microprocessor Each cycle creates a new microprocessor implementation Each cycle creates a custom circuit (Ascenium instruction) representing hundreds to thousands of conventional instructions Programmed using ANSI-standard programming languages (e.g., C/C++) Tens to 100s of times faster than a general-purpose microprocessor Dynamically reconfigurable microprocessor limitations There are none on the market today Until Ascenium, no one has figured out how to “program” a dynamically reconfigurable circuit VCs don’t understand it 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 Microprocessors x86 AMD, Intel, Transmeta, Via ARC ARC ARM ARM MicroBlaze Xilinx MIPS MIPS Nios Altera PowerPC IBM, Freescale SPARC Sun Tensilica Stretch, Tensilica Old stuff Everyone This is not a comprehensive list of microprocessors, but it’s the list that I’ve chosen to deal with and these are the companies that I’ll talk about. At least I’ll talk about a subset of these companies. If there’s one you want to know about that you don’t see here, ask about it in the minus ten minutes we’ll have for Q & A. 27 March 2017 Nick Tredennick Nick Tredennick

Microprocessor Applications
The Rise of Reconfigurable Systems 27 March 2017 Microprocessor Applications Supercomputers Workstations and servers PCs Embedded systems Automobiles Cameras Cell phones Game players MP3 players Set-top boxes Here’s my list of categories for applications. There are supercomputers, workstations and servers, PCs, and embedded systems. 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 Computer Markets Here’s an illustration of the relative sizes of the computer markets. If you are building x86 microprocessors for the Windows market, your total available market is perhaps 180 million—if you are Intel. If you are building microprocessors for the workstation and server markets, your TAM is a few million units. If you are building microprocessors for the supercomputer market, your TAM is a few thousand units. Notice that I didn’t include the embedded microprocessors here. You’ll see why in the next chart. 27 March 2017 Nick Tredennick Nick Tredennick

Microprocessor Markets
The Rise of Reconfigurable Systems 27 March 2017 Microprocessor Markets Here’s what happens to the chart when embedded microprocessors are included. Suddenly, the PC market, which dominated the previous chart, looks like someone stepped on it. Servers and workstations are barely a measurable patch and I didn’t even bother to show supercomputers. The market for embedded microprocessors is measured in billions. You see this chart infrequently because the embedded microprocessor market is split among many manufacturers and because margins are low. This chart is in units. If I had made the chart in dollars, the PC bar would be as large as the embedded bar. Margins are much, much higher in PCs, in workstations and servers, and in supercomputers. 27 March 2017 Nick Tredennick Nick Tredennick

Computer Microprocessors
The Rise of Reconfigurable Systems 27 March 2017 Computer Microprocessors x86 AMD Intel Transmeta Via Proprietary IBM Freescale Sun I divide computer microprocessors into two groups: x86 and proprietary. Among the x86 manufacturers, Intel dominates the market. AMD has made the right strategic decisions to capture market share and it has executed well to put itself in a good position. Intel blundered by diverting its resources to Itanium while AMD extended the x86 with 64-bit instructions. AMD’s second important strategic decision was implementation of the Hypertransport bus, which facilitates multiprocessor implementations. Transmeta is struggling for position. Via is in a good position as a low-cost supplier to emerging markets and it is also well-positioned for embedded applications (more on this later). IBM and Freescale make proprietary PowerPC microprocessors, which go into IBM computer systems and into Apple’s computers. The PowerPC microprocessor is also offered for embedded applications and it is licensed to Xilinx for its Vertex programmable logic devices (more on these later too). Computer applications for PowerPC are a small market that is likely to stay small. Sun is circling the drain. In 1991, I wrote a paper for Microprocessor Report. The title of the paper was “MIPS and Sunset.” In that paper, I forecast the demise of the workstation companies as they would be squeezed out by rising PC performance. I predicted that Sun would be the last to go because it had locked the most customers into a proprietary operating system from which it would be the most difficult to migrate. 27 March 2017 Nick Tredennick Nick Tredennick

Embedded Microprocessors
The Rise of Reconfigurable Systems 27 March 2017 Embedded Microprocessors x86 AMD, Transmeta, Via ARM ARM PowerPC IBM, Freescale Old stuff Everyone Triscend (Xilinx) For embedded microprocessors, I see four categories: x86, ARM, PowerPC, and “old stuff.” In the old stuff category are all of the thousands of microprocessors that have evolved for various embedded systems over the years. For embedded x86, I believe that Via has the best position followed by AMD. Transmeta is a distant third. ARM, of course, dominates high-end embedded applications because it dominates cell-phone implementations. Its position seems secure and its future seems assured. I have doubts that I hope we can get to. The old stuff is too fragmented to be profitable enough to discuss here. I’ve listed Triscend here because it was a very interesting company. Triscend offered microcontrollers with a fixed microprocessor core and a standard set of peripherals and with programmable logic and software that let the application engineer “drag and drop” peripherals to create custom microcontrollers that would perfectly fit any application. The microprocessor cores on its chips were great choices: the 8051 and ARM. I saw this as an opportunity for Triscend to consolidate the microcontroller market. It didn’t happen. ARM offered to buy Triscend for $13 million in a move that I liked because it would have given ARM a way to get into embedded applications. That didn’t happen either as Xilinx stepped in with an offer double ARM’s, but still less than half of what had been invested in the startup (as a seed-round investor, I noticed). It looks like Xilinx didn’t want ARM peddling programmable logic. I don’t expect much from Triscend’s products under Xilinx’s care. 27 March 2017 Nick Tredennick Nick Tredennick

Configurable Microprocessors
The Rise of Reconfigurable Systems 27 March 2017 Configurable Microprocessors ARC ARC Ascenium Ascenium MIPS MIPS Nios Altera Tensilica Stretch, Tensilica These are the companies that make configurable microprocessors. ARC, MIPS, and Tensilica make what I call design-time configurable microprocessors. That means that logic designers will have to be involved in designing new instructions to fit the application. These new instructions are implemented in a custom chip for that customer. ARC and Tensilica have sophisticated development software for this process; MIPS is a late comer and does not have sophisticated tools. Nios is also a configurable microprocessor, but I will talk about it later with the PLD microprocessors. Ascenium is a startup company whose products will benefit greatly if the configurable microprocessor market develops, so it’s a discussion for the future. That leaves Stretch. Stretch uses a Tensilica microprocessor core and implements that in a chip with programmable logic. The chip is generic at manufacture and it is customized in the field. I call this run-time configurable as opposed to design-time configurable and I expect Stretch to be the vanguard of the new category and the first really successful configurable microprocessor company. The next step after run-time configurability is dynamic reconfiguration, which is left for Ascenium. 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 PLD Microprocessors Altera Nios (soft) Xilinx MicroBlaze (soft) PicoBlaze (soft) PowerPC (hard) At one time, Altera offered ARM hard cores on its Excalibur FPGAs. Altera does not offer hard cores in either its Stratix chips or in its Cyclone chips. Altera’s strategy, which I think is a good one, is to concentrate on its soft-core Nios microprocessor. Xilinx, in contrast, offers hard-core PowerPC microprocessors on its high-end Vertex chips and it offers soft-core MicroBlaze and PicoBlaze microprocessors too. 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 Microprocessor-like DSPs Network processors Specialty processors I don’t think the NPU business is dead; it’s just a dead end for all but a couple of companies. Here’s a nutshell view. I see the NPU companies repeating the mistakes of the workstation companies—they are mostly (Intel is an exception) creating high-end, proprietary architectures (this includes EZChip). In the workstation business, the build-for-performance (and lock your customers in) strategy was thoroughly defeated by Intel’s build-for-volume strategy. Build for volume and performance follows. The performance of the x86 rose faster than that of the workstation processors because the x86 had a rapidly growing market that supported rapid design evolution. The PC came up in performance to encroach on the workstation markets. As the PC encroached, the workstations retreated to higher-end applications until they painted themselves into corners with few customers. The same will be true in the network processor business, the companies that build for volume at the low end (a few, including Intel) will overtake the companies (most of them) that build for the high end. A second strike against the NPU market is the onset of the dumb network. Routing should be done at the ends of the network; packets should travel through the middle of the network without being manipulated. This decreases the need for high-speed packet processing and concentrates the routers at the low-speed ends of the network. PCs can do much of this routing and there’s room for a company or two with high-volume, standard-architecture routers at the low end. 27 March 2017 Nick Tredennick Nick Tredennick

ASICs & Microprocessors
The Rise of Reconfigurable Systems 27 March 2017 ASICs & Microprocessors This is the way the market segments look to me on a plane of performance and problem size. The ASIC market offers the ultimate in performance in a market of about $30 billion. The microprocessor offers the ultimate in flexibility in a market of about $40 billion. 27 March 2017 Nick Tredennick Nick Tredennick

The Rise of Reconfigurable Systems 27 March 2017 ASICs & Microprocessors Configurable microprocessors are a new category that straddles the two markets. 27 March 2017 Nick Tredennick Nick Tredennick

The Rise of Reconfigurable Systems 27 March 2017 ASICs & Microprocessors Over time, I expect configurable microprocessors to push out enough area to be their own category at the expense of both ASICs and microprocessors. 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 Semiconductor Trends Value PCs outsell leading-edge PCs Mobile applications emerge Design emphasis shifts from cost performance to cost-performance-per-watt Value transistors outsell leading-edge transistors Transistor performance overshoots many applications Increasing demand in emerging economies Foundry strength grows Emerging mobile applications will shift the design emphasis from performance to cost-performance-per-watt. Since the invention of the integrated circuit, the transistor hasn’t been good enough, so all applications helped pay for the next-generation process. Now, just as with the personal computer, the transistor is good enough for many applications. Those applications will no longer help pay for semiconductor improvements. The vast majority of applications aren’t leading-edge, performance-oriented systems. Most applications want adequate performance and minimum cost. The industry has been focused on Moore’s law because the transistor wasn’t good enough; in the future, what engineers do with transistors will be more important than how small they are. There’s still plenty of opportunity. 27 March 2017 Nick Tredennick Nick Tredennick

27 March 2017 Consequences Rise of mobile applications New non-volatile memories Rise of foundries Rise of soft (IP) cores Horizontal fragmentation of integrated device manufacturers Rise of non-volatile FPGAs Rise of reconfigurable systems Growing market for embedded microprocessors Tethered: traditional role Mobile: supervisory role Foundries, which are demand driven, can service the range of applications. Foundries also amortize fixed costs across a broad customer base. Programmable logic devices are generic in manufacturing and are customized in the field—the best case to accommodate escalating fixed costs for semiconductor masks and equipment. Soft cores support design portability and design productivity. They also enable generic components to be customized in the field. As a consequence of the change in orientation from speed to cost-performance-per-watt, makers of microprocessors and digital signal processors will find that their speed orientation isn’t a good fit for cost-performance-per-watt applications. An increasing number of design wins will go to the makers of application-customized programmable logic devices. This is a topic in its own right. Fabs are good enough: foundries are a good investment, but equipment makers aren’t. 27 March 2017 Nick Tredennick Nick Tredennick

Industries in Transition
The Rise of Reconfigurable Systems 27 March 2017 Industries in Transition Automotive Analog ► Digital Mechanical ► Electrical Isolated ► Connected Telecom Copper ► Optical, Wireless Biomedical Wet labs ► Bioinformatics Film/video Consumer Tethered ► Untethered Computers Desktop ► Embedded There’s an ongoing and depressingly fruitless search for “the next killer app.” It’s time to give up the search; its view of the future is too narrow. Instead, let’s look at opportunities offered by industry transitions. Here are examples. The automotive industry is in transition from analog to digital, from mechanical to electrical, and from isolated to connected. The film and video industry is in transition from analog to digital and from isolated to connected. The consumer-products industry is in transition from analog to digital, from tethered to untethered, and from isolated to connected. (The consumer-products industry also offers high-growth opportunities in supplying standard appliances to emerging economies.) The biomedical industry is in transition from analog to digital and from wet laboratories to bioinformatics. The telecom industry is in transition from copper to untethered, from copper to fiber, and from electrical to optical. The computer industry is in transition from desktop to embedded. These transitions will completely transform these industries. Transitions in these industries will create large markets for MEMS, for electrical and electronics components, and for computers and software over the next few years. The first big transitions, from machine-room computers to personal computers and from wired telephones to cell phones, were viewed as “killer apps.” But they were the pioneering applications that opened the floodgates for the general transition from analog to digital. 27 March 2017 Nick Tredennick Nick Tredennick

Reconfigurable Systems Emerge

Similar presentations

Presentation on theme: "Reconfigurable Systems Emerge"— Presentation transcript:

Similar presentations

About project

Feedback

Log in

Auth with social network:

Reconfigurable Systems Emerge

Similar presentations

Presentation on theme: "Reconfigurable Systems Emerge"— Presentation transcript:

Similar presentations

About project

Feedback