65-nm Processes Target Next-Gen High-Speed And Low-Power ASICs

A pair of 65-nm CMOS processes lets designers implement ASICs with up to 120 million gates. Developed by IBM, these processes deliver twice the number of gates of the company's previous 90-nm offerings. These copper- and low-k-based processes also can achieve a gate density of 615 kgates/mm² while keeping the power dissipation to less than 5 nW/MHz/gate.

One process fits low-power designs, while the other suits high-performance applications. Both can set up voltage islands to control power consumption and use multiple-threshold-voltage library building blocks to balance performance and power requirements. They also incorporate strained-silicon structures that can improve the transistor performance by as much as 40% over non-strained devices (see the figure).

Design support for both processes includes tools that provide integrated noise, power, and timing methodologies to enable first-pass success. The tools use statistical techniques in timing and optimization to address process variations. This lets designers correct errors that might previously have gone undetected and re-engineer the chip throughout the design flow, achieving a more robust design.

Standard-cell design libraries, multiple I/O families, embedded SRAM and DRAM, embedded processors, and a wide range of packaging options support the processes. A broad selection of additional intellectual property (IP), including high-speed serializer-deserializers (SERDES), also supports the processes.

The low-power process (Cu-65LP) can trim leakage currents by as much as 30 times more than IBM's previous 90-nm ASIC. Transistor leakage is as little as 0.01 nA/µm when the chips are powered by a 1.2-V supply.

The per-gate power dissipation of 5 nW/MHz is just slightly lower than the active power of the 90-nm process. Yet the overall average power for an ASIC implemented in the Cu-65LP process is significantly lower thanks to the reduced leakage current. A design kit for the low-power process is now available, but volume production of chips designed for the process won't be ready until the first quarter of 2007.

Pushing the high-performance limits, the Cu-65HP process targets high-frequency performance-driven applications in networking, communications, data processing, and storage networks. It can deliver about a 20% performance increase over IBM's 90-nm process while operating from a 1-V supply.

Gate delays ranging from 6 to 9 ps are possible, while power dissipation is just 4.5 nW/MHz/gate. The design kit will be available in the fourth quarter of this year, with volume production of Cu-65HP devices projected for the third quarter of 2007.

Contact the company for development and production costs.