Saturday, January 15, 2011

AnXplorer: New generation analog and RF circuit optimization

Analog and RF design is rapidly growing. Risk of respin remains high with predominantly manual design process.

AnXplorer from AgO can automate key manual optimization and simulation activities. High performance optimizer produces designs meeting objectives and constraints. Benefits include design team productivity, yield improvement, as well as reduced respins.

Design methodology has changed little over the years. Manual, iterative design with many SPICE runs are common. In the AgO design methodology, AnXplorer automates device resizing and SPICE runs.

AnXplorer uses industry standard inputs, which are compatible with existing design flows. It generates optimized and centred netlist that meets or exceeds objectives.

It works with existing environments. Supported simulators include Cadence Spectre, Synopsys HSpice, Legend Design Technology MSim and Mentor Eldo. It also provides multi-threading support. AgO’s AnXplorer uses Red Hat Enterprise Linux 4 OS.

Why the need?
Do remember that analog and RF ICs are everywhere. Analog and RF circuits are key to many systems. Analog/mixed-signal is also growing at 13 percent CAGR. Besides, analog design is expensive.

Analog circuits account for~2 percent of total transistor count, account for 20 percent of total IC area and 40 percent of total design effort. In fact, it is responsible for 50 percent of design respins. It requires specialist skills. Now, analog designers are in short supply. Design automation for analog lags that for digital design. Tasks are said to be labour intensive.

Analog design further deteriorate in smaller process nodes. For instance, costs of an analog design increase with finer geometries. Design cost per IC cited as going from ~$4 million with 0.18µm to ~$46 million with 65nm (Source: Jim Hogan, “Escape from analog Alcatraz”, DAC 2006 Interoperability Breakfast).

Optimization approach
AgO has a three stage optimization approach. One, it does feasibility analysis using DC operating point analysis. Two, it does global optimization using single PVT corner. And three, it does local centering using all PVT corners. It also supports simulation- and equation-based optimization. Besides hierarchical design objectives, trade off analysis with exploration database is possible, as is multi-threading support.

Multi-algorithmic optimization approach
Multi-algorithmic strategy is controlled by an expert system. Device rules are stored in knowledge base to ensure safe DC operation. This includes intelligent handling of devices in saturation or linear and recognition of current mirrors, level shifters, etc. Designer know-how is used in objectives and constraints.

Enhanced sizing rule – automatic detection
AnXplorer will evaluate and automatically identify the following building blocks. Tough optinization problems are tackled, especially with smaller geometries circuit optimization problems are not well-behaved and have multiple local minima. Many existing optimization tools use convex or gradient optimization. AnXplorer has successfully found the global minimum in the presence of many local minima for tough benchmarks such as Rastigin’s function.

Equation-based optimization
AnXplorer has different modes of operation – graphical mode and command line mode. Equation-based approach supports early design exploration. Analog systems can be described by parameterised functions. AnXplorer can optimize arbitrary functions. Functions are defined in standard C.

Productivity improvement
Engineering efficiency should be improved by at least 5×. Highly manual and iterative sizing of circuit elements is automated. AnXplorer also enables experienced staff to focus on choosing circuit topologies.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.