The Z80 Lives!The Z80 and its derivatives continues to be tremendously successful. Here's the current state of the art.
Published in Embedded Systems Programming, March 1992
For novel ideas about building embedded systems (both hardware and firmware), join the 27,000+ engineers who subscribe to The Embedded Muse, a free biweekly newsletter. The Muse has no hype and no vendor PR. It takes just a few seconds (just enter your email, which is shared with absolutely no one) to subscribe.
By Jack Ganssle
I get really nervous reading the trade press. Magazines will have you believe that the only viable processor for even the simplest applications is a 50 Mhz 486 or a screaming RISC machine. How many designers are using these CPUs in their embedded designs?
At the Embedded Systems Conference Andy Rappaport made a convincing argument that one day we'll have infinite MIPs for zero dollars. Until that time I remain convinced that smaller CPUs, those that address the price sensitivity of so many embedded products, will dominate new designs.
Personally, I prefer 8 and 16 bit CPUs. They're cheap, they're cheap to interface to, and they are easy to work with. The processor itself costs about 1/50 of a 68030 or 80386. Remember that 16 and 32 bit machines will have 2 or 4 ROM and RAM chips per word, compared to 1 for an 8 bit processor, greatly increasing memory costs.
In fact, an 8 bit processor exactly matches the bus width of the vast majority of memories and peripherals. An 8 bit bus uses memory efficiently (handling byte data in a 16 bit word is wasteful). Simpler circuitry, with fewer components, can be used.
16 and 32 bit busses are often used to move more data in parallel. A wider instruction word, handled in a single machine cycle, means more complex operations can be accomplished in less time. Larger addresses can be used efficiently, giving the machine access to more memory. The benefit is speed at the penalty of more cost.
EDN's annual Microprocessor survey came out in December. Though I didn't count the number of current micros, the sheer bulk of the survey is an impressive testimonial to the huge number of different choices available to the embedded designer.
I feel four processor families can satisfy most embedded designs. I'll leave out 4 bit choices, as these applications are often so specialized that a custom or semi-custom chip is often the most cost effective solution. My choices are the 8051 family, the Z80 family, the 80186 (and V-Series) families, and the 68000 and its derivatives.
Occasionally I'll do an article about members of each family. This month's choice is the Z80.
Current Z80 Technology
The Z80 is essentially unchanged from the original version introduced in the mid-70s. Now CMOS versions are common, and clock rates have skyrocketed. Zilog's newest offering runs at 20 Mhz. This speed is no panacea - it comes at the price of expensive ROMs and RAMs. Still, even at 6 to 10 Mhz the Z80 has respectable performance. It would be a terrible Windows machine, but is more than adequate for an awful lot of embedded systems.
The Z80 itself is attractive due to its low cost and wide availability. It's pretty easy to find Z80s for less than a buck. Tools are everywhere. Though its index instructions are a little crude, it is a far better C machine than, say, the 8051.
The Z80 in isolation would be a dead-end line. What makes it interesting is the high integration derivatives CPUs spawned by the architecture. All four of the processor families I mentioned earlier exist in multiple proliferation versions. A processor by itself is useful; one integrated with a number of on-chip peripherals is compelling.
The 64180 is a Hitachi-supplied Z80 core with numerous on-chip "extras". Zilog's version is the Z180, which is essentially the same part. As of this writing, Zilog sells parts running at speeds to 15 Mhz.
The 64180/Z180 microprocessor integrates many of the functions traditionally assigned to peripheral circuitry onto a single chip. The designers picked an architecture compatible with the Z80, giving Z80 users a completely software compatible upgrade path. Old Z80 designs can be converted to the 64180 with essentially no loss in software investment. New designs will benefit from the processor's low cost, powerful instruction set, minute power consumption, and high level of integration.
64180 peripherals include two 16 bit counter/timers, programmable refresh and wait state generation, a pair of DMA controllers, and three serial ports with on-chip baud rate generation. Toss in a little memory and some parallel I/O and you have a complete computer.
Like the Z80, the 64180 supports 16 bit I/O addresses. Whenever an IN A,(C) or OUT (C),A instruction is executed, both the Z80 and 64180 issue the contents of C as the low order port address, and the contents of B as the high 8 bits of the address. Unlike the Z80, for which few users ever needed more than an 8 bit address, 16 bit port addresses are crucial to the chip's operation.
The 64180's internal peripherals are all accessed via reads and writes to I/O ports. Each device is assigned a set of unique port addresses, whose upper 8 bits is always zero. In other words, to read or write an internal I/O device your program must issue an IN or OUT with a 16 bit port address, and that address must be less than 256.
The 64180 reserves 64 I/O addresses for dealing with the internal I/O addresses. On reset, these addresses are from 0000 to 003F. It is possible to set the starting address of the I/O registers to any of four values: 0000, 0040, 0080, and 00C0. This is particularly useful if the internal register conflicts with a port in the user's external hardware.
CPU Bus Programmability
The 64180 permits dynamic selection of several bus parameters, including memory wait states, I/O wait states, and refresh rate.
Like the Z80, the 64180 automatically generates refresh cycles to keep dynamic memories alive. Now that statics are available inexpensively in large configurations, refresh just degrades processor performance. On the 64180 the programmer can select the refresh rate. Refresh can be altogether disabled if desired.
The Refresh Control Register enables refresh, sets the rate, and optionally adds a wait state during refresh for slow memories. The fastest possible refresh is set at reset with the wait state enabled. While nicely Z80 compatible, this results in a significant performance degradation. If refresh is not needed, it should be disabled.
The 64180's refresh controller generates an 8 bit refresh address (the Z80 generates a 7 bit address). Unfortunately, 1 Mb dynamic RAM chips all need a 9 bit refresh address. It's pretty easy to place a divider on bit 8 of the address to generate refresh bit 9.
The 64180 has two independent 16 bit timers that both count the CPU's clock. Each can be programmed to generate an interrupt when the count reaches zero. The timers are automatically reloaded with an initial value at a count of 0.
Each timer has a 16 bit data register that can be read or written to set/examine the current count value, and a 16 bit reload register that is loaded into the data register when a count of 0 is reached.
Using the timers is quite simple. The data registers are set to an initial value, the reload registers are set to the reload value (usually the same number as the initial data register contents), and the proper mode is set in the Control register.
The 64180 contains three serial devices. Two of these are traditional asynchronous UARTs (ASCI 0 and ASCI 1). The third is a clocked port for synchronous applications.
Each ASCI (Asyncronous Communications Interface) supports full duplex, 7 or 8 data bits, 1 or 2 stop bits, all forms of parity, some modem control signals, and separately controllable baud rate generation. Interrupts can be generated on a received character, data error, or transmit buffer empty condition.
The clocked Serial I/O port (CSI/O) is a simple synchronous port that supports half duplex 8 bit communications. It is not complete enough to drive SDLC or other complex synchronous links, but can drive the serial bus used by the 8051 and other processors.
Like the Z80, the 64180 can run in one of three interrupt modes (0, 1, and 2). In addition, it supports a number of new interrupt sources.
Four external sources of interrupts are available. NMI and INT0 correspond with the Z80's nonmaskable NMI and maskable INTR inputs. These behave in exactly the same manner as on the Z80. INT0 can run in mode 0, 1, or 2, depending on the selection made via the IM instruction.
Two new external inputs, INT1 and INT2, always function in mode 2 (vectored), regardless of the specified IM mode. When asserted, the CPU branches to a table to fetch the address of the processing routine.
Similarly, all interrupts from internal I/O devices are always handled in mode 2.
On the Z80, all mode 2 vectors are taken from a table in the memory page specified by the setting of the I register (which can be accessed by the LD I,A instruction). The interrupting device must put the low order 8 bits of the vector address on the bus during the interrupt acknowledge cycle. This remains the case on the 64180 for the external INT0 interrupt source. However, the INT1, INT2, and internal devices work somewhat differently.
A new register (IL) specifies bits 5, 6, and 7 of the vector table for INT0, INT1 and internal interrupts. The upper 8 bits of the vector are still given by the I register, just as on the Z80. Note that INT0 does not use the IL register in forming the vector address, to maintain Z80 compatibility.
INT1, INT2, and the internal interrupts are assigned fixed offsets from the start of the table formed by I and IL.
The 64180 processor runs every Z80 instruction exactly as a Z80 does, though a bit faster. All registers are handled the same, all flags are identical (which is not quite the case between the 8085 and Z80), and all addressing modes are the same. This makes the transition to the 64180 for Z80 programmers quite easy. A few new instructions are useful in some applications.
The SLP instruction puts the 64180 in an ultra low power mode. Sleep mode can only be exited by the receipt of an interrupt, or by reset. SLP can also be used for ultra-low latency response to interrupts, as any interrupt during a SLP just exist sleep mode without executing an interrupt cycle.
An 8 bit unsigned multiply instruction gives a 16 bit result. Using only 17 T-states, it is a very fast way to multiply two small numbers. It takes any register pair (BC, DE, HL, or SP), multiplies the upper register by the lower one, and returns the result in the pair.
The four versions of the TST instruction all perform non-destructive bitwise ANDs of the argument against the accumulator. TSTIO takes as its argument an 8 bit I/O port address. The upper 8 bits of the 16 bit port specifier are automatically set to 0.
One of the biggest advantages of a 16 bit processor over an 8 bit one is its ability to address a large address space. Most 8 bit computers are limited to 64k of memory, whereas a 16 bit CPU generally supports 1 to 16 Mb. The memory may be cumbersome to access, as in the 8088 series, but it is available. A large memory space is important for big data structures and programs.
The MMU is the mechanism by which the 64180 can access an address space larger than 64k, but still maintain compatibility with the Z80 instruction set. All instructions, in all modes, always issue 16 bit addresses. The MMU converts every one of these 16 bit addresses to 20 bits.
The memory connected to the 64180 is its "physical memory"; it is the entire universe of memory available to the processor. The memory that can be addressed with any one map, or configuration, of the MMU is called the "logical" address space. Every address generated by a user's program is a logical address. The MMU's role is to translate these logical addresses into physical ones. If, for example, the user's program tries to reference location 0, this is a reference to logical address 0. The actual, physical, address accessed will be a function of the mapping of the MMU. It may indeed use physical address 0; on power up, the MMU translates every logical address to exactly the same physical address (which simulates the Z80).
Every address reference made by a program is passed through the MMU before being placed on the CPU's pins. If the address matches a range previously programmed into the MMU, then the MMU will add an offset to that address, forming the physical address. Since the offset is a 20 bit number the device can access up to 1 Mb of memory.
The 64180's MMU lets the user segment the memory into up to three different areas. For instance, references to logical addresses 0 to 3FFFh could access physical addresses 0 to 3FFFh, references to 4000h to 7FFFh could access physical addresses 10000h to 13FFFh, and the last half of the logical space could access physical addresses 40000h to 47FFFh. Virtually any combination is possible.
A more complete discussion of the MMU is beyond the scope of this article. Many vendors have helpful application notes. Some compilers even automatically support MMU, remapping it dynamically to let you code huge programs.
Other Z80-Like CPUs
The 64180/Z180 is not the only high integration derivative of the 64180. Both Zilog and Toshiba sell a number of Z80-like devices with differing mixes of peripherals and memory. The range of processors grows daily.
Both Toshiba and Zilog sell the 84013 and 84015, which are Z80 cores with conventional Z80 peripherals integrated on-board. While not as popular as the 64180/Z180, these processors are natural migration paths for current users of Z80s wishing to reduce systems costs.
Both the 84013 and the 84015 include one SIO (Z80-specific Serial I/O), and one CTC (again, a Z80-specific timer part). The SIO and CTC are functionally identical to the discrete chip SIO/CTC used in so many older designs. In addition, the newer parts include a watchdog timer and clock generator.
The 84015 comes (unfortunately, to my way of thinking) in a 100 pin quad flat pack. To take advantage of the extra pins the vendors added in a Z80-like PIO (parallel I/O) port, with 16 parallel lines and 4 handshaking likes.
Zilog and Toshiba sell a number of other Z80-like parts, which are of more interest to our hardware brethren than to programmers. The 84C50 is a Z80 with 2k of RAM. The 84C01 is a Z80 with integrated clock circuits.
Toshiba's TMP 90 line are properly not Z80 parts, as the machine code instruction encoding is different than the Z80. However, it does maintain source compatibility, so could be considered Z80-like. The TMP 90 line includes dozens of family members, each having a different mix of peripherals and on-board memory.
Typically, a TMP 90 part comes with ROM (up to 32k), RAM (up to 1k), 4 to 6 analog to digital channels, one or more timers, parallel I/O, and DMA. Pulse width modulation channels (worthy of an article in themselves) and zero crossing detectors are also included.
The TMP 90 parts are especially nice because of their on-board ROM and RAM. This is a cheap way to make high volume single chip systems. Some of the parts come in PROM versions, making prototyping much more economical.
Does anyone really know what processors most designers use? I believe this is a mystery.
Chip vendors track volume sales of CPUs to determine which ones outsell the others. They're happy to sell 100 million CPUs to one customer. This tells us nothing about the number of designers working with a chip. A thousand applications being produced in hundred quantities amounts to few chip sales, but lots of engineering.
What processors do you use? Why? Let me know and I'll pass the information on to your fellow readers.