The Cyclone™ II FPGA is fabricated on TSMC’s 90-nm process and uses 12” (300 mm) wafers. The FPGA uses low-dielectric-constant (low-k) (k = 2.9) dielectric and copper for the metallization layer so that it can take full advantage of the potential speed and dynamic power consumption benefits offered by the advanced feature sizes. A total of eight all-copper layers are used that enable very high area efficiencies through increased wiring density.
Choosing the right fabrication partner is critical in order to take advantage of right fabrication process for a silicon product. The right process choice balances production worthiness and cost predictions over the product lifecycle against technical requirements. In this way, cost effective and advanced devices can be offered, whilst fully meeting delivery commitments.
TSMC Market Leadership
TSMC is the silicon foundry market leader with 53 percent market share and a year-on-year revenue growth of 26 percent, which hit USD $5.9B in 2003 (see Figure 1). With this leadership position, TSMC can invest more than its rivals so it can offer the most advanced capabilities; this is particularly important considering a new fabrication facility’s cost approaches USD $3B. TSMC’s advanced capabilities are paired with the industry’s highest yield, which is ultimately a metric for the quality of production. This investment has already been proven with the success story of Altera’s 130-nm-based product rollout (i.e., within 15 months from first samples, TSMC was able to supply Altera with enough processed wafers to allow shipment of over 2 million Cyclone devices with no delivery issues).
Figure 1. 2003 Worldwide Dedicated Foundry Industry by Revenue
Altera & TSMC Partnership
Altera’s strong partneship with TSMC has paid off in terms of on-time delivery performance, quality, and cost—but it also brings other not so obvious benefits. Over the last five years, FPGAs have shifted from using mature process technologies to leading-edge process technologies. FPGAs have the broad market appeal and volume to justify using the latest technologies, but they also benefit most from the area reduction and performance benefits, and this makes them an increasingly desirable choice over ASICs—even at higher volumes. Producing FPGAs with the latest process technologies is also beneficial to the foundry since they are the lead vehicle for a new process.
Trying to find a functional error due to a processing defect in a microprocessor is very difficult. However, because of an FPGA’s regular row and column architecture and programmable structure, finding a functional error is relatively simple. This enables rapid learning on the new process node, which leads to a faster yield ramp. Because of this, producing FPGAs with the latest processes is benefitial for both Altera and TSMC that provides close collaboration in new process technologies development.
90-nm Development & Low-k Dielectric
Early research and development into 90-nm products was essential for a timely 90-nm product rollout. TSMC and Altera jointly started work on 90-nm process technology in May 2001; since which time there has been an extensive effort in developing new architectures, materials, and processes for building 90-nm devices. Since May 2001 multiple test chips have been jointly developed to evaluate different properties of the 90-nm process.
Throughout the development process, TSMC ran monthly wafers containing large SRAMs to accurately evaluate yield against process targets. With so many parameters affecting yield and performance, this type of continuous assessment is essential for improving quality in a reasonable timeframe. TSMC’s 90-nm low-k, copper metal process has been fully qualified since July 2003, and Altera’s first 90-nm product (Stratix II EP2S60 device) is fully functional and shipping to customers. Debugging new processes on test chips rather than on products has been fundamental to meeting delivery commitments. A crucial component for success is that the manufacture of the 90-nm test chips and process qualifications have taken place in the same fabrication plant that is producing Cyclone II FPGAs; this removes the risk associated with changing lines. Early process technology development work with TSMC and the correct product rollout strategy ensure that delivery of Cyclone II sample devices is fully on track.
Figure 2 shows copper interconnect at 90nm.
Figure 2. Copper Interconnect at 90nm
Low-k has been a significant and recent challenge within the industry. Altera decided not to employ low-k dielectric in first-generation Stratix and Cyclone devices for two reasons: it wasn’t essential for good performance at 130 nm, and, more importantly, it wasn’t considered to be a mature enough material for reliable production.
For Stratix II FPGAs and Cyclone II FPGAs, Altera selected Applied Material’s Black Diamond material, offered by TSMC, which is a chemical vapour deposition (CVD) dielectric for 90-nm based products because of its superior manufacturability.
What Is Important
In addition to good product design, there are other elements in making a FPGA successful, including:
Strong Partnership for Success
Altera partners with TSMC exclusively. TSMC is the foundry leader not only in terms of revenue and technology but also yield, which ultimately demonstrates superior quality.
Right Time
Altera and TSMC debug new processes using test chips. Altera employs the newest processes for high-end, fully featured products first; Stratix II FPGAs are the first Altera 90-nm FPGAs. By the time Cyclone II FPGAs are available in production, the 90-nm process at TSMC will be an established process. Cyclone II FPGAs will be in production when TSMC attains the high yield critical to a low-cost product.
On Time
Altera achieved a flawless execution at 130 nm using this winning strategy. This winning strategy will work again for Cyclone II FPGAs with rapid volume availability which is especially important for a low-cost, fast time-to-volume FPGA. Technology is the cornerstone of digital CMOS, but should not be pursued at the expense of good economics. Altera and TSMC together have the right balance of advanced technology and production know-how, which takes the risk out of choosing Cyclone II FPGAs.
