IC sockets fall into two distinct categories:

• Sockets for Production: There are many applications in electronic equipment.
• Sockets for Test & Burn-In (T&B): T&B of IC packages, PCBs, and other test.

These two application areas have distinctly different characteristics:

• Design: Production sockets are designed for one-time installation, packaging automation. T&B sockets are designed for thousands of insertion cycles and rigorous test parameters.
• Pricing: Production sockets = high volume/low cost; T&B sockets = low volume/high cost.
• Customer Base: Production sockets = electronic OEMs and CMs; T&B = IC manufacturers and other test

Sometimes production sockets are used in test applications, but rarely is the opposite true. This is because the production sockets are lower in cost.

The lineup of suppliers is different in this area because most T&B socket suppliers are specialists in the field. Some companies do both, but with the production versus test ratio of 20:1 in the connector market, there isn’t as much room in the T&B segment.

Bishop & Associates pegs the worldwide IC socket market in the $1.5 billion range.

Given the large influence IC design has on connector applications and usage, and with T&B socket designs arguably on “the leading edge,” T&B production may be a strategic imperative for the connector manufacturer.

Many companies may not see it that way, or they believe the lower volumes and higher engineering content in T&B overrides the leading-edge theory. I know of many debates on this subject over the years. There is even a question about the sales aspect, with production connectors perhaps not requiring the same level of design-in, field expertise, or specialization. Nonetheless, these segments are important to the marketplace and are well served by many companies; both product types make major contributions to the ergonomics, reliability, productivity, and connectivity of the electronics industry.

Test & Burn-In Sockets:

This market is characterized by many challenging design and application parameters:

• The re-use, 10-100,000s of times without degradation.
• A minimal footprint on IC or test pads.
• A reliable connection to challenging shapes and soft surfaces such as solder balls.
• Electrical parameters, such as very high edge rates.
• Environmental parameters, such as 70-125ºC operating temperatures.
• Designs that allow multiple ICs to be tested in a magazine or other fixture.
• The trend toward chip scale and flip chip presents extreme dimensional challenges.
• The trend toward wafer scale test changes the ball game in some applications.
• Test costs keep rising, in some cases presenting a barrier to future test strategies.

IC test will be around for many years to come but is constantly going through periods of change with users trying to minimize escalating test costs:

• Packages change from DIP and SIP, to TSOP, BGA, BGA, QFN, and C4/Flip Chip.
• Pad or lead spacing continues to shrink from 2.54mm and above to < 0.5mm.
• BGA packages have gained huge popularity, with their particular test challenges.
• IC performance continues to rise with Moore’s Law well into the Gb range.
• Burn-in becomes a challenge due to cost, particularly in high volume DRAM and flash applications.

Singulated test sockets are included in the electronic connector market. Wafer probes, a $400-$500 million market, are not. These connectors are closest to silicon technology trends, which include some of the most pressing requirements in the industry:

• Multi-Gb/s speed.
• High density approaching the limits of discrete connector technology.
• Low voltages and noise margins.
• Ever higher power and current loads.

Test sockets are used for package and die verification, device characterization, failure analysis, and debug. Once the package/device is characterized the manufacturer generates a final test program, which is used with the test socket/contactor during large volume final test.

The key component of a test socket is long life, i.e., high insertion count. Not all devices require high performance; burn-in sockets can be used as test sockets because they are inexpensive. In general, a long-life/high-performance test socket will cost 10 times as much — or even greater — than a burn-in socket, and deliver 300,000-500,000 insertions before a rebuild. Key factors in high-performance test connectors are lower voltages and faster edge rates. With lower voltages, connectors are getting noisier by comparison to sockets with higher voltages, resulting in crosstalk, etc. With microprocessors dropping below 2 volts, we are down to what was considered the noise level only a few years ago. With 5V TTL logic and 20-80% as the unknown state, there is a low-end level of 1V. Today that 1V is very close to the high. Therefore, what the noise ceiling was a few years ago is now the Logic 1 floor. Edge rates are faster, with tighter timing window requirements. Even if a part only runs at 250MHz, edge rates are important, because that part may be syncing with another part running at 3GHz. Thus, when the part is clocked, it has to change state as though it were a 3GHz part. If it didn't, it would miss the time window, or cause a latent state during the 2-3 clock cycles to change state. This requires test sockets that do not delay or distort the signal.

There is also a need for a lower force connector that does not crack thinner die. This is especially true of larger devices which include BGA, LGA, and PLCC. The die is made thinner to increase performance and reduce cost, but the pin count is rising—up to 10,000 I/O on the roadmap for some server ASICs and CPUs. Even the current AMD Opteron Socket F is now over 1,000 I/O (1207). Other areas include an increase in strip testing—mainly for lower speed mixed signal, small QFN and SO-type devices. Strip applications are dominated by pogo pin solutions and need to withstand up to 1 million touchdowns. Finding a contact probe or metal that will handle continuous solder contacting (solder balls) without becoming inter-metallic is an issue, if not a roadblock. Solder buildup increases contact resistance, which changes all the DC and AC parametrics and also increases yield fallout. Current solutions require frequent changes of the test probe, which increases retest costs.

Obviously, T&B sockets carry many challenges:

1. Density
2. Speed
3. Contact Design
4. Cycle Life
5. Engineered Materials and Design Engineering Content
6. Application Engineering

Production Sockets

Production sockets have also evolved over the years, perhaps even more so than test and burn-in sockets. IC sockets used to include primarily DIP, SIP, SIMM, and PGA, with some proprietary CPU sockets. These are still around, as are many smaller-volume machined sockets. But now the big players are DDR SO-DIMM (notebook) and DIMM (desktop/server DRAM memory sockets), and the various iterations of Intel/AMD microprocessor sockets such as PGA, PGA, and LGA.

There are also proprietary LGA sockets for RISC/MIPS processors in servers and workstations produced by IBM, SGI, Sun, and others. This category is quickly merging with X86 systems, which use the PGA and LGA varieties.

Arguably, the many different memory cards — particularly SFF cards — are in effect an IC socket, even though they aren’t reported that way. The IC packages that house SD, Micro-SD, memory stick, and other SFF cards, are essentially an IC package with pads. They mate to a memory card socket, which is typically a linear slide-in connection, as opposed to pin and socket, pin grid, or land grid array. These applications, as well as the Pentium/Opteron varieties, are produced in huge volumes. One of the latest is the 1207 pin Socket F for AMD Opteron processors. The socket at right is an LGA which fits at the base of a heat sink stack on Opteron motherboards and has a steel backing plate. This arrangement allows the IC package to be built with pads rather than leads, which reduces cost and improves electrical performance.

Why Socket?

There are many trade-offs between using a socket and direct attachment of a component. BGA packages “upped the ante” by providing a surface tension-aligning package with a low profile. It costs more to socket an IC than to direct attach, particularly now with the BGA dynamic. Through-hole or surface mounting can cost less than a quarter of a cent per pin, while socket costs range from a half cent to 10 cents or more per pin. The reasons ICs are socketed vary, but the bottom line is flexibility and, in the end, applied system cost.

Reasons sited for socketing include:

• Avoidance of tariffs.
• Protection of expensive components from damage or theft in a global supply chain.
• Insurance against faulty designs in new systems.
• Flexibility in motherboard design and final assembly.
• Upgradeability, expansion and reconfiguration after the sale.

As pin counts have gone up, so have cost and reliability challenges for IC packages. By far the most elegant current design (outside of no package at all, such as flip chip) is the Organic Land Grid Array (OLGA). This incorporates a low-cost, high-performance substrate and eliminates leads from the IC package, replacing them with copper-nickel-gold (Cu-Ni-Au) pads that can be solder balled (BGA) or socketed (LGA). A BGA package is not socket-friendly because of solder ball plasticity. This can be accommodated by an interposer, however, more cost effective is the BLGA socket design, which has a solder ball attachment on the socket itself and provides a low, self-aligning mated height.

Dynamics of the Production Socket Market

There are two broad applications areas in the production socket market.

The computer industry is the primary market segment for production IC sockets. The main applications include the CPU (Intel Pentium/Xeon/Itanium, AMD Athlon/Sempron/Opteron, and various high-end RISC/MIPS processors) and memory sockets (DDR/Dual In-Line Memory Modules). There are also plastic leaded chip carriers (PLCC) and plastic quad flat packs (PQFP). It is understandable that sockets are used for all of these reasons, given the open system architecture of PCs; their global supply chain for motherboards, daughtercards, other component parts; and for subsystem and final assembly. This market is subject to significant changes, which are discussed below.

The second broad area is ... everything else. Applications ranging from industrial equipment to vending machines to RFID security systems to central office telecom gear use some sockets. Often these are garden variety DIP sockets. In other cases, many different specialized machined sockets are applied in small to midsize volumes. Often, more than one IC part can be used to program equipment, which the system is designed to accept via a socket. This part of the market is extremely diverse and is slowly evolving. With the digital convergence of embedded computer functions in all types of equipment, more systems, including consumer products, are using CPU, DIMM, SODIMM, and PLCC types, plus huge quantities of memory cards.

Dynamics in the Computer Market

Years of evolution in proprietary, customized computer hardware is giving way to standardization. Broadly speaking, RISC/MIPS and other OEM workstation, server and even super-computer designs are shifting over to X86—not all, but definitely in the volume market.

Within this dynamic, which is most closely associated with Intel and Microsoft, is another factor: the rise of AMD and open-source Linux operation systems. For example, there is an inexorable drive for lower cost, which promotes the use of standard systems. This includes component parts, and the market volumes already achieved in the mainstream of the computer market: in motherboards, Pentium/Xeon/AMD processors, DDR memory, and IC sockets. In servers, rack and blade server systems, which look more like PCs, are rapidly gaining ground against big box systems. At the very high end, massively parallel systems composed of hundreds of PCs are challenging proprietary super-computer designs in performance, and at considerably lower costs. Many of these systems have 2, 4, 16, or more CPUs (and sockets).

Thus, down in the food chain, PGA, BLGA and DIMM sockets are on the rise, while custom high-end LGA sockets are being truncated. This is not entirely a “commodity story” with low cost, “cheap” components, because computer performance is going up, requiring ever more sophisticated socket designs. Among them:

• Surface mount BGA attachment, which is not trivial.
  • 1,000 pin area array sockets with attendant co-planarity, mating force, and other issues.
• Mated height < 3mm to accommodate mobile and thin blade systems
  • 1GHz < 1H performance with low crosstalk, unheard of just a few years ago.
• Ability to operate in high peak current and thermal loads as chip performance escalates.

Bishop & Associates Summary:

• The world IC socket market is approximately $1.5 billion.
• This does not include the huge market for memory card sockets (SD/MMC, memory stick, etc).
• IC sockets are broadly viewed as falling into either T&B or production socket categories.
• The market split in dollars is 20:1 in favor of production sockets.
• T&B sockets are typically high cost, have thousands of insertion cycles, and have specialized suppliers.
• Emphasis in production sockets is cost, design for assembly, and IC/equipment performance.
• T&B sockets are evolving with IC performance. Challenges include miniaturization, electrical performance, pressures to reduce escalating test costs, and wafer scale test.
• Production sockets have two main application areas: computer equipment and everything else.
• In computers, PGA, BGLA, DIMM, and PLCC dominate, with high-end systems giving way to x86. This means that more x86 sockets will be used in the future, and perhaps less custom LGA.
• Socket performance is going up with IC technology, and costs are coming down, so that x86 CPU and memory sockets are continually challenged in performance and productivity.
• There will always be a trade-off between using a separable socket interface and direct attachment of an IC package. There are many reasons why OEMs will continue to specify sockets, not the least of which is the massive trend toward global assembly of subsystems, coupled with an increasing need for system flexibility.

This subject is more extensive than this space allows. If you have a particular interest, let us know and we will try to address it in a future article!

John MacWilliams
Senior Consultant and Analyst, Bishop & Associates Inc.
John MacWiIliams has been in the electronics industry for more than 40 years. His main areas of experience have included: U.S. competitiveness programs, market research studies, authored articles, field sales and management, product marketing management, strategic marketing, new product planning, venture development, advertising and media relations, direct sales, manufacturers representative, distribution sales management, and international marketing. MacWilliams has worked with AMP, Diceon Electronics, TRW, and IRC in marketing management positions. Prior to joining Bishop & Associates, MacWilliams served as the group director of marketing and new product planning for AMP.

MacWilliams graduated from Lehigh University with degrees in business management and engineering.