MicroVAX II (1985)

MicroVAX (formally MicroVAX II, "code name" uVAX) was DEC's first single chip VAX microprocessor, and the first microprocessor in the industry with virtual memory management. Bob Supnik was the project manager for the CPU, Larry Walker for the FPU. Dan Dobberpuhl was project lead for the CPU, Bob Simcoe for the FPU. Rich Witek did the CPU microarchitecture; Bob Supnik wrote the CPU microcode.

MicroVAX was intended to establish an industry standard for 32b processing by creating a single-chip VAX microprocessor. To reach this goal, much of the complexity of the VAX architecture -- h_floating, commercial instruction set, compatibility mode, console, standard peripherals -- was tossed overboard. In fact, the initial subset definition included a different memory management scheme, primarily to limit the scope of systems built around MicroVAX. When the impact of this change on VMS was fully understood, MicroVAX reverted to standard VAX memory management.

In addition to subsetting the architecture, MicroVAX cut down the implementation complexity as far as possible. There was no support for caching; the processor ran in lockstep with memory. The translation buffer (TLB) was only 8 entries and contained both system and process entries. All external communications -- to memory, to the FPU, and to support chips -- was done over a single multiplexed bus, with very simple timing and protocols. Despite all these simplifications, MicroVAX ran at 85% to 90% of the performance of the VAX-11/780, because of its faster memory access time (400ns vs 1600ns).

The MicroVAX CPU and FPU were implemented in DEC's 3u double-metal NMOS process (ZMOS) and ran at 5Mhz. Although faster bins were possible, most system designs locked processor and memory timing and could not use faster chips.

Name Number Size Transistors Comments
MicroVAX CPU DC333 353x358 125,000 sites The MicroVAX CPU is a high performance, single chip 32b microprocessor that implements a compatible subset of the VAX architecture. Key features include:
  • High performance
    • 32b internal and external data path
    • Pipelined architecture
    • Instruction prefetch
  • Subset VAX architecture
    • Sixteen 32b general purpose registers
    • 175 instructions
    • 21 addressing modes
    • 6 data types
  • VAX memory management
    • 4GB virtual address space
    • 64MB physical address space
    • Demand paging
    • Memory protection
    • Four privilege modes
  • Vectored multi-level interrupts (15 software, 7 hardware)
  • Industry compatible external interface
  • Single +5V supply

Power: 2.5W.

MicroVAX FPU DC337 339x272 34,000 The MicroVAX FPU is a high performance, single chip floating point processor for the MicroVAX CPU. Its key features are:
  • High performance
    • Implements all floating point operations
    • Accelerates integer multiply and divide
  • f_, d_, and g_floating point format support
  • Full VAX floating point instruction set, including ACBf, EMODf, POLYf
  • Arithmetic error checking and reporting
  • Single +5V supply

Power: 2.5W.

MicroVAX systems shipped in May, 1985 and were an instant hit. Their low price, good performance, small form factor, and full VMS compatibility made MicroVAX systems appealing for all kinds of departmental applications, as well as OEM and embedded use. Another milestone was the shipment of the VAXstation 2000 in 1987, which brought VMS to the desktop. The VAXstation outsold all other workstation models that year.

Personal Narrative

MicroVAX was not my idea; indeed, it wasn't DEC's idea. In the summer of 1981, Zilog approached DEC about building a single chip, subset VAX to set an industry standard for 32b computing. This idea was so intriguing that Roy Moffa formed a MicroVAX Program Office to investigate the business implications. As manager of the Advanced Development Group, I joined the team to provide technical guidance. The Program team like the idea of a subset VAX and an industry standard but wanted a different partner than Zilog. While they talked to all the "major" semiconductor players, I worked with Dave Cutler and Dick Hustvedt to define the subset. We quickly arrived at a definition that left out d_ and h_floating point, the commercial instructions, the console, and all the standard peripherals, and that changed system space from virtual to physical (this limited the size of systems that could be built). I also worked with Dan Dobberpuhl (then a consultant to the Advanced Development Group) to develop a baseline of what a chip might look like, how big it might be, how much it might cost, etc. We used existing architecture and componentry from V-11's IE chip to construct a model of MicroVAX in DEC ZMOS, as a comparison for proposals from industry partners.

Unfortunately, no proposals were forthcoming. The Program Team was so frustrated that it proposed adopting an industry chip (the National NS32032, which hadn't shipped) and porting VMS to that. I counterproposed that DEC build the chip itself, based on the feasibility work done to date. After a short but heated debate, the company's top management approved MicroVAX, and I was told to go build it, and quickly.

That was easier said than done. Most of DEC's chip designers were busy finishing J-11 or working on V-11. In order to build MicroVAX quickly, I begged and borrowed resources and components wherever possible. The core microarchitecture of the CPU chip, as well as the main data path, the microsequencer, and parts of the instruction decoder, were taken from V-11; the control store, memory management unit, and bus interface unit were new. The FPU was a direct lift of the J-11 FPA. The CPU team was assembled from consultants (like Dan Dobberpuhl and layout lead Tim Thrush), newcomers (MicroVAX was Rich Witek's first architectural assignment, and John Beck's first NMOS design), and new hires; the FPU team was equally eclectic in its origins. MicroVAX was allocated 2.5 VAX-11/780's for design use; that's about half the computing power in a modern optical USB mouse.

The chip taped out at the end of January, 1984, only 21 months after the initial proposal, and 19 months after project start. The fab processed the lot in record time; but, horrifyingly, the first pass parts didn't function: the control store didn't work reliably. Heroic debugging work led by Rich Witek enabled the testers to overdrive the control store from test inputs. This way, 100 of the 175 instructions, and most of the chip logic, was shaken down. Pass 2 taped out, amidst high hopes, in June, 1984.

It didn't work either. Another bug in the control store again prevented stand-alone operation. Laser trimming got the chip running on a test bench and then in a real system; VMS booted successfully in August, 1984. Nonetheless, it would take several more passes to get all the circuit bugs out of the chip. Indeed, DEC's ability to simulate large circuits at the time was so limited, and the correlation between models and actual process so poor, that eventually the design team had to guess about the right values for fixing the control store. Fortunately, they guessed right.

MicroVAX was presented at the 1984 International Solid State Circuits Conference. It won the company's outstanding product award for 1986.


Updated 24-Feb-2008 by Bob Supnik (simh AT trailing-edge DOT com - anti-spam encoded)