System and method of computer operating mode clock control for power consumption reduction

[0002] This is a continuation of application Ser. No. 09/121,352 filed Jul. 23, 1998, U.S. Pat. No. 6,079,025; which is a division of application Ser. No. 08/767,821 filed Dec. 17, 1996, abandoned; which is a continuation of application Ser. No. 08/460,191 filed Jun. 2, 1995, abandoned; which is a continuation of application Ser. No. 08/285,169 filed Aug. 3, 1994, abandoned; which is a continuation of application Ser. No. 08/017,975 filed Feb. 12, 1993, U.S. Pat. No. 5,396,635; which is a continuation of application Ser. No. 07/908,533 filed Jun. 29, 1992, abandoned, which is a continuation of application Ser. No. 07/532,314 filed Jun. 1, 1990, abandoned.

[0003] A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.

[0004] The present invention relates to computers and particularly to methods and apparatus for power management in computers, particularly in battery-powered computers.

[0005] The major parts of computers include a central processing unit (CPU), input/output (I/O) devices such as display screens, keyboards, modems, printers, disk drives and the like, and storage (memory).

[0006] The CPU communicates with the I/O devices, with the storage and otherwise operates with addresses defined within the computer address range. Typically, addresses for I/O devices are within an I/O address range. Addresses for execution of programs without I/O reference typically are within a memory address range. Similarly, that portion of memory allocated for display is within a video memory address range.

[0007] Computers function to execute application programs such as word processing, spreadsheet and data base management programs. Typically, the computer and the application programs are under the control of a software operating system that manages the different system parts and resources including some I/O devices. For example, during the execution of an application program when the CPU wishes to check to determine if any key has been depressed on the keyboard, the CPU through a subroutine call to the operating system requests the operating system through execution of a subroutine to perform a key-actuation detection task. Since the operating system performs many such tasks, the operating system has a detailed knowledge of many activities within the computer. However, under some circumstances, application programs bypass the operating system and directly address I/O devices. Typically, each I/O device is assigned an I/O address within an I/O address range. For application programs which-directly address I/O devices without operating system calls, the operating system is not immediately aware of I/O activity. With such complex operation in computers, the task of power conservation is difficult.

[0008] The need for power conservation is well known in battery-powered computers and must be performed in a manner that does not interfere with the operation of the computer or impede users from interacting with the computer during the execution of application programs.

[0009] Conservation of power has been utilized for some parts of battery-powered computers but has been ignored for other parts of such computers. In general, power consumption is distributed in battery-powered computers among the major parts of those computers. One part with significant power consumption is the central processing unit (CPU). Another part is the input/output (I/O) devices such as display screens, keyboards, modems, printers, disk drives and the like. Still another part with significant power consumption is storage (memory).

[0010] Prior art attempts at conserving power have employed screen blanking which reduces the power to the display screen when the screen has not been used for some period of time. Typically, a timeout circuit senses changes in screen information and, if no change has occurred for a predetermined timeout period, the backlight to the screen is turned off for power reduction. While screen blanking is effective in reducing power for the display screen, no reduction results in power to the driver circuitry for the display, to the CPU, or to other parts of the computer. Furthermore, when the screen is blanked, the computer cannot be used until reset.

[0011] Other prior art attempts at conserving power consumption have focused on disk drives because the power consumption of rotating magnetic disks is high. Disk drive manufacturers have employed various schemes for reducing the power consumption of the disk drive. While such power consumption schemes are effective for the disk drive, no reduction results in power to the CPU or other parts of the computer. Computers without disk drives, such as small “notebook” computers, have no need, of course, for the conservation of power in a disk drive.

[0012] In order to extend the battery life of portable computers and to manage power in computers, there is a need for improved power management methods and apparatus in computers, particularly for power management that can be extended to many different parts and conditions of the computer.

[0013] The present invention is a method and apparatus for power management in a computer. The computer typically includes as hardware a central processing unit (CPU), storage (memory) and I/O devices and includes as software an operating system adapted to control the computer during application program execution.

[0014] The power management method and apparatus causes the computer system to enter the power conservation mode after sensing inactivity by a software monitor or by a hardware monitor.

[0015] The software monitor monitors the activity of the operating system or other software in the system. The software monitor typically is a software module linked, for example, to the operating system at boot time for monitoring subroutine calls to the operating system.

[0016] The hardware monitor monitors the hardware to detect inactivity. The hardware monitor typically is circuitry for detecting inactivity independently from the software. For example, the hardware monitor senses predetermined address ranges, such as an I/O address range and a video memory address range, and monitors the activity of addresses by the CPU to addresses within these ranges. If no data transfers occur within the specified address ranges for predetermined periods of time, then a power conservation mode is entered to conserve power in the computer system.

[0017] By using both a software monitor and a hardware monitor, the power management unit determines exactly when to enter into power conservation mode without sacrificing system performance.

[0018] In the software monitor, inactivity is determined by detecting how many “active” or “idle” function calls an application makes within some time period. In the IBM PC DOS environment, the activity status is checked, for example, no less frequently than every 50 milliseconds. There are 256 IBM PC DOS function calls and, in principle, each is labeled as “idle” or “active” and each is assigned a corresponding positive or negative number. A positive number is assigned to an “active” function call and a negative number to an “idle” function call.

[0019] The power management software monitor forms an activity measurement as a running total of the function call numbers as the function calls are made. Whenever a function call is made (either active or conservation), the power management software monitor algebraically adds the function call number to the accumulated value and determines whether the system is to remain in the active mode or be switched to the conservation mode by comparing the magnitude of the accumulated value with a function call threshold.

[0020] The function call threshold for determining activity is a variable depending on the computer system speed. To prevent the system from oscillating between the active and conservation mode due to minor changes in system activity, hysterisis is provided by using active and conservation function call thresholds. The accumulated total for the activity measurement is reset after it reaches the active threshold going in one direction or the conservation threshold going in the opposite direction as the case may be.

[0021] The active and conservation thresholds are typically unequal so that the entry and exit from conservation mode is biased. For example, in order to have the system enter the conservation mode quickly and thereby to reduce power consumption, the active threshold is set with a number greater than the number for the conservation threshold.

[0022] In one embodiment, functions that require immediate attention are assigned numbers large relative to the active and idle thresholds so that a single occurrence of the function call will force the accumulated count over the active threshold and thus force the system to be in the active mode. The hysterisis effect can be bypassed by forcing the power management unit into active mode without changing the activity count. In this case, the next idle function call will bring the system back to idle mode.

[0023] If the software monitor or the hardware monitor indicates inactivity, the power management unit enters the conservation mode. The conservation mode has multiple states which provide different levels of power conservation.

[0024] A first state, called a DOZE state, is entered after sensing inactivity by the hardware monitor for a first period of time. A second state, called a SLEEP state, is entered after sensing inactivity by the hardware monitor for a second predetermined time where the second predetermined time is greater than the first predetermined time. A third state, called a SUSPEND state, is entered after sensing inactivity by the hardware monitor for a third period of time greater than the first and second time periods.

[0025] Another state is OFF which turns off all power for the computer under predetermined conditions.

[0026] During periods of inactivity, power consumption is reduced in different ways, for example, by reducing clock speeds or removing clocks, and/or by removing power, and/or by controlling the refresh frequency to memory.

[0027] In accordance with the above summary, the present invention achieves the objective of providing an improved power management method and apparatus.

[0028] The foregoing and other objects, features and advantages of the invention will be apparent from the following detailed description in conjunction with the drawings.

[0029] FIG. 1 depicts a block diagram of a computer with the power management unit of the present invention.

[0030] FIG. 2 depicts a block diagram of the power management unit of the FIG. 1 system.

[0031] FIG. 3 depicts a detailed block diagram of the hardware for the power management unit of FIG. 2.

[0032] FIG. 4 depicts a state diagram depicting the multiple states associated with the power management unit of FIGS. 1, 2 and 3 as determined by the hardware monitor.

[0033] FIG. 5 depicts a representation of operation for various states as a function of the activity measurement.

[0034] FIG. 6 depicts a state diagram depicting switching to conservation mode (DOZE or SLEEP state) operation under control of the software monitor.

[0035] FIG. 7 depicts a state diagram depicting the sequencing which forces to the ON state during an activity window period under control of the software monitor.

[0036] FIG. 8 depicts a representation of operation for a spreadsheet application program.

[0037] FIG. 9 depicts a representation of operation for a word-processing application program.

[0038] FIG. 10 depicts a representation of operation for a windowing application program.

[0039] Computer System—FIG. 1

[0040] In FIG. 1, computer 3 is typically a small, battery-powered computer such as a “notebook” computer. The computer 3 includes a CPU 4, a CPU bus 5, a plurality of I/O controllers 6-0, . . . , 6-n where “n” is a constant equal, for example, to 7. Connected to the controllers 6-0 through 6-n are plurality of peripheral devices 7-0, . . . , 7-n, respectively. The controllers and peripheral devices 6 and 7 typically include a keyboard, a display, a hard disk drive, a modem, a printer, and similar devices. Each of the controllers 6-0 through 6-n connects to the conventional computer bus 5.

[0041] Also connected to the bus 5 is the memory, which in one particular embodiment is DRAM random access memory 11. The memory 11, when of the type requiring refresh, is refreshed with *RAS and *CAS lines 29 under control of the PC controller 13 which provides *PCRAS and *PCCAS signals on lines 30 to power management unit 15 including a hardware monitor 79 and a software monitor 80. The I/O devices are separately powered through switch unit 22 and switches 22-0, . . . , 22-n by the VCC power from power supply 9 which receives power either from the battery 10 or an AC source 14. Power supply 9 is of a conventional type which supplies a low battery signal LB, a low-low battery signal LLB, and an AC power signal ACPWR to power management unit 15.

[0042] The computer 3 typically includes as software an operating system adapted to control the computer system and to control operations during application program execution. Computer 3 functions to execute application programs such as word processing, spreadsheet and data base management programs. Computer 3, during the execution of application programs, is under control of a software operating system. The operating system manages the different system parts and resources including the I/O devices 6 and 7. For example, during the execution of an application program when the CPU wishes to check to determine if any key has been depressed on a keyboard I/O device, the CPU 4 through a subroutine call to the operating system requests the operating system to execute a subroutine to perform a key-actuation detection task. Since the operating system performs many similar calls to the operating system, these calls represent detailed information about many activities within the computer system.

[0043] In FIG. 1, the computer 3, through the CPU 4, issues control and address signals on the bus 5 which define the overall computer address range for computers including the sets of address ranges for all of the memory, I/O and other devices connected to the bus 5. Whenever any of the peripherals 7-0 to 7-n are to be accessed for data to be transferred over the bus 5, the address of the corresponding I/O controller 6-0 to 6-n (either by unique address lines or unique address lines in combination with control lines) specifies the addressed one of the I/O controllers 6 and corresponding peripheral 7.

[0044] Similarly, memory 11 has locations addressed by a set of addresses on bus 5 within a memory address range. Some of the addresses in the range of addresses for memory 11 are typically allocated and reserved only as a set of video memory addresses. Whenever the video memory region 8 of memory 11 is to be addressed, address appears on bus 5 within the set of video memory addresses.

[0045] The computer system of FIG. 1 includes a power management unit 15 having a software monitor 80 and a hardware monitor 79 for monitoring activity of the computer system. The power management unit 15 is connected to the bus 5 to sense activity, using hardware monitor 79, on the bus 5 and is connected to the CPU 4 (executing the operating system and the software monitor 80), the power supply 9, the memory 11 and PC controller 13 for controlling power management.

[0046] The power management unit 15 of FIG. 1 operates to cause the computer system to enter the power conservation mode after sensing inactivity by the hardware monitor 79 or by the software monitor 80 and to enter the active mode after sensing activity or other conditions.

[0047] The hardware monitor 79 monitors the hardware to detect inactivity. The hardware monitor 79 typically is circuitry for detecting inactivity independently from the software and the software monitor 80. For example, the hardware monitor 79 senses predetermined address ranges, such as an I/O address range and a video memory address range, and monitors the activity of addresses by the CPU to addresses within these ranges. If no data transfers occur within the specified address ranges for predetermined periods of time, then a power control mode is entered to conserve power in the computer system.

[0048] The software monitor 80 monitors the activity of the operating system or other software in the system. The software monitor 80 typically is a software module linked, for example, to the operating system at boot time for monitoring subroutine calls to the operating system.

[0049] By using a software monitor 80 and a hardware monitor 79, the power management unit 15 decides exactly when to enter into power conservation mode and active mode without unnecessarily sacrificing system performance.

[0050] The power conservation mode includes a number of activity states. A first state, called a DOZE state, is entered after sensing inactivity for a first period of time by the hardware monitor or when an idle threshold is exceeded as determined by the software monitor. A second state, called a SLEEP state, is entered after sensing inactivity by the hardware monitor for a second predetermined time where the second predetermined time is greater than the first predetermined time or when the activity measurement sensed by the software monitor exceeds the idle threshold. A third state, called a SUSPEND state, is entered after sensing inactivity for a third period of time greater than the first and second time periods. Another state is OFF which turns off all power for the computer under predetermined conditions.

[0051] After having entered one or more of the activity states of the conservation mode, the power management unit switches back to the active mode when activity is sensed by the monitors.

[0052] Power Management Unit—FIG. 2

[0053] In FIG. 2, a block diagram of the power management unit 15 of FIG. 1 is shown. The power management unit includes a hardware monitor 79 (including an activity monitor 16 and a timer unit 24), a software monitor 80, a state control unit 23, a power control unit 17, a clock control unit 18, and a refresh control unit 20. The hardware monitor 79 (using activity monitor 16) analyzes the address activity on the system bus 5 to provide activity information used to control power management. The timer unit 24 times the activity information sensed by the monitor 16. The state control unit 23 controls the changes among different power consumption states to achieve power management.

[0054] The power control unit 17 controls the switches 22-0, . . . , 22-n of FIG. 1 as a function of the activity sensed by activity monitor 16 and the state determined by state control unit 23.

[0055] The clock control unit 18 controls the distribution of and/or the frequency of the CPU and other clocks as a function of the activity sensed by the activity monitor 16 and the state determined by state control unit 23.

[0056] The refresh control unit 20 controls the refresh of the RAM memory 11 of FIG. 1 at a rate which is determined by the activity sensed by the activity monitor 16 and state control unit 23.

[0057] The power management unit (PMU) 15 is provided to manage power and reduce, over time, the overall power consumption of computer 3. This management is accomplished using an activity monitor 16 to detect periods of system inactivity. During periods of inactivity, power consumption is reduced by reducing clock speeds or removing clocks through clock control unit 18, and/or by removing power through power control unit 17, and/or by controlling the refresh frequency through refresh control unit 20. Standard and slow refresh DRAM support is provided by refresh control unit 20. Inputs are provided to the power management unit 15 which will allow power on or off commands from external sources such as a pushbutton, modem ring indicator, or read-time-clock (RTC) time of day alarm.

[0058] Hardware Monitor Generally—FIG. 3

[0059] Referring to FIG. 3, the power management unit (PMU) 15 includes the hardware monitor 79 (activity monitor 16 and timer unit 24) which is designed to operate with minimal system requirements and without software support. Power management occurs in response to the hardware monitor independently of any operating system (DOS) or application program support.

[0060] In FIG. 3, the PMU 15 has its own power-on reset signal (*RESET) which is produced by a VCC power detector 71, separate from any other reset signal of computer 3, and upon initial power-on, the registers of the power management unit 15 are initialized to preestablished default values to provide basic functionality without need of any software.

[0061] While the hardware monitor 79 and the power management unit 15 are provided FIG. 3 as a hardware embodiment, a software embodiment of the hardware monitor 79 is described in the program listing of TABLE 1. Using the program listing of TABLE 1 executing in the CPU 4, power management, using a software embodiment of a hardware monitor, occurs under program control.

[0062] In accordance with the operation of the hardware monitor 79, a predetermined set of address ranges on bus 5 is monitored by power management unit 15 as part of the power management operation. For example, the predetermined set of address ranges monitored for power management typically includes all of the I/O address range, that is, the addresses of the I/O controllers 6-0 through 6-n and the video memory address range for the video memory locations 8 within the memory 11. Of course, other address ranges can be added to or used as the predetermined set for power management. The set of address ranges including the video memory and the I/O address ranges has been found to provide excellent information for controlling power management.

[0063] The hardware monitor 79 senses the activity of addresses on the bus 5. Whenever addresses within the predetermined set of addresses are not present on the bus 5 for predetermined time periods, the power management unit 15 responsively switches power consumption states and controls the consumption of power by different parts of the computer 3.

[0064] The power management unit 15 has four main operating states, namely, ON, DOZE, SLEEP, and SUSPEND, and a fifth state which is OFF. The five power management states, under control of the hardware monitor 79, are shown by the state diagram of FIG. 4. The activity monitor 16, external inputs (EXT, RESET), and the timeouts of timer unit 24 generally control the transitions between states in the state control unit 23 as shown in the state diagram of FIG. 4. The CPU 4 of FIG. 1 may also command the PMU 15 to enter any state. The commands from the CPU 4 typically derive from execution of the software monitor 80, but may derive from other CPU 4 commands.

[0065] In FIG. 3, each of the four active states (not OFF) has an associated PWR register which indicates in one embodiment which of eight power control outputs VP[0 . . . 7] on lines 33 will be active during the state. More generally, any number, (n+1), outputs VP[0 . . . n] can be employed. The PWR registers in power control unit 17 are PWRON register 57, PWRDOZE register 58, PWRSLEEP register 59 and PWRSUSPEND register 60 as shown in FIG. 3. A power control multiplexer 76 selects the eight outputs from one of the registers 57 through 60 corresponding to the current state on STATE lines 34 from unit 23, and these eight outputs drive the VP[0 . . . 7] power control outputs from EXOR unit 35. Also, the CPU 4 of FIG. 1 can write, under program control, to any of the PWR registers 57 through 60 to control which of the I/O devices 6 and 7 are powered at any time.

[0066] To turn an I/O device on, the corresponding bits in the PWR registers 57 through 60 for the state(s) in which they are to be on is typically high. The POLARITY register 61 specifies the actual polarity of each output VP[0 . . . 7] required to turn the associated one of the switches 22-0, . . . , 22-n on and thereby supply power to the I/O devices 6 and 7. The default value of the POLARITY register is 03h, which implies a logic low to turn on VP[2 . . . 7], which will typically control logic switches 22 with low-true output enables (for example, switches 22 typically include a PNP transistor in the VCC line from power supply 9) and high to turn on the LCD, VP[0], and EL backlight, VP[1], power. The value of the VP[0 . . . 7] bits just prior to the polarity control by EXOR 35 may be read back through the OUTPUT register 62 to CPU 4 over bus 5.

[0067] The system clock oscillator signal CLKI is connected to the CPU Clock Control block 49 to produce the CLKOUT. From there CLKOUT, as controlled by PMU 15 and control block 49, drives CPU 4. The CLKOUT clock can be stopped for static CPU's, or reduced automatically by a divisor specified in the CLOCK field of control register 53 during DOZE and SLEEP states. CLKI is passed through unchanged to CLKOUT in SUSPEND state.

[0068] Detailed implementations of the various monitor, control and logic blocks of FIG. 3 will be clear from the following detailed description. Additionally, a software embodiment of the hardware monitor 79 including logic and control functions equivalent to those in the hardware embodiment appears as the Program Listing of TABLE 1.

[0069] Software Monitor Generally

[0070] The software monitor 80 of FIG. 2 includes a power management software module linked into the operating system, for example, during boot up time. One embodiment of the module appears as the program listing of TABLE 2.

[0071] The software monitor 80 monitors all the function calls to the operating system. Every time an idle function call is made, the activity measurement, AC(t), is incremented and then checked against thresholds. The incrementing is algebraic by the amount of D_a, a positive DOS call number, or D_i, a negative DOS call number.

[0072] If the activity measurement, AC(t), is below the idle threshold, T_H, and the system is in the active mode, no action will be taken. However, if the activity measurement, AC(t), is above the idle threshold, T_H, the power management software will check the current system status and if in the active mode, will switch to the conservation mode.

[0073] The activity measurement, AC(t), is given by the following Eq. (1): $\begin{array}{cc}\sum _{a,i}{\lfloor {|}^{\lceil }{D}_a\left(t\right)+D_i\left(t\right)=AC\left(t\right)|\rfloor }^{\rceil }& \mathrm{Eq}.\left(1\right)\end{array}$

[0074] where,

[0075] D_a(t)=Active DOS call numbers as a function of time

[0076] D_i(t)=Idle DOS call numbers as a function of time

[0077] AC(t)=Accumulated Activity Count of DOS call numbers as a function of time, that is, activity measurement

[0078] While all of the interrupts of the operating system may be assigned a D_aor D_ivalue the following, for example in the following CHART 1.

[0079] Using the values in CHART 1, each time an interrupt 16 (I16) occurs, the software monitor increments AC(t) by +12 and each time I10 or I8 occurs the software monitor increments AC(t) by −25. The value of AC(t) is shown for one example of operation in FIG. 5.

[0080] Referring to FIG. 5, the value of AC(t) as a function of t is shown. In the example of FIG. 5, the first eight values of t find keyboard polling occurring by the I16 interrupt so that +12 is added to AC(t) for each of the first eight values of t. In FIG. 5, at t=8, the timer interrupt I8 occurs and subtracts −25 from the AC(t) value. Thereafter the keyboard polling continues until the value of AC(t) reaches 128, the value of T_Hin the example of FIG. 5. At t=12 in FIG. 5, AC(t) is reset, for example, to 0 when the computer system enters the conservation (idle) mode. At about t=20 in FIG. 5, which may include a long time duration generally indicated by the broken line at about t=15, video interrupt I10 becomes active and starts to add −25 to the AC(t) value until at about time t=35 the value of AC(t) reaches the −256 value of the threshold T_L.

[0081] When the value of AC(t) is above T_H, then the software monitor is operative to switch the computer system into the conservation mode. Whenever AC(t) is in below the threshold T_L, the software monitor is operative to switch the computer system back to the active mode.

[0082] The example of FIG. 5 is only for purposes of representing the manner in which AC(t) is incremented as a function of the positive and negative interrupt call numbers. Of course, other counting methods may be employed. In the program of TABLE 2, after the T_Hvalue of +128 is reached, the counter is reset to +256 and each value of Da decrements the count until the threshold TL is reached at 0.

[0083] The operation which occurs when the value of AC(t) exceeds the threshold T_H, is explained with respect to the flowchart of FIG. 6.

[0084] In FIG. 6, the value of D (either Da or Di), the interrupt number value, is added as indicated in Eq. (1) to form the accumulation value of the activity measurement, AC(t). This accumulation is indicated by the oval marked D in FIG. 6.

[0085] Next, the value of AC(t) is compared with the threshold T_H. If the value of the summation in Eq. (1) is not greater than the threshold, T_H, then the N no choice is made the loop repeats so that the next value of D is added to the AC(t) activity measurement. For example, in FIG. 5, this activity continues until approximately t=12 in FIG. 5.

[0086] In FIG. 5, at about t=12, the activity measurement AC(t) equals or exceeds the threshold T_Hand hence the Y output of the comparison connects to the SLEEP state detector. If already in the state, then the Y output will force the computer system to remain in the SLEEP state. If not in the SLEEP state, then the software monitor will force the computer system into the DOZE state.

[0087] Note that the FIG. 6 operation will force the computer system into the DOZE or SLEEP state as long as the activity measurement AC(t) exceeds the threshold T_H. When the threshold T_Hhas been exceeded, AC(t) is reset and remains reset until another activity event, Da or Di, occurs. In FIG. 5, for example, this occurs at about t=20 when AC(t) begins to count toward T_L.

[0088] In addition to the comparison of the activity measurement AC(t) against the upper threshold T_H, the software monitor 80 also compares the value of the activity measurement against the lower threshold T_L. This comparison is represented by the flowchart of FIG. 7.

[0089] In FIG. 7, the oval represents the incrementing of the activity measurement AC(t) in accordance with Eq. (1). After each incrementing of the activity measurement, the value of AC(t) is compared to determine if it is less than or equal to T_L. If not, then the N output of the comparison continues the incrementing of the activity measurement for each new value determined in accordance with Eq. (1).

[0090] If the activity measurement AC(t) is less than or equal to T_L, then the Y output of the comparison connects the operation to the activity window comparison.

[0091] If AC(t)≦T_Land AW(t)≦T_aw, then the FIG. 7 operation switches to the ON state.

[0092] If AC(t)≧T_H, then test sleep state.

[0093] where,

[0094] T_H≧K₁

[0095] T_L≦K₂

[0096] T_H=Idle Threshold

[0097] T_L=Activity Threshold

[0098] K₁=128

[0099] K₂=−256

[0100] Combined Hardware Monitor and Software Monitor Operation

[0101] If the system is in ON state and AC(t) is greater than or equal to T_H, the power management software monitor will bring the system into DOZE state. If the system is already in DOZE or SLEEP state, no further action will be needed. Similarly, the activity count, AC(t), will be decremented every time an active function call, Da, is made. The activity count is then used to compare with the active threshold. If the count is higher than the active threshold, T_H, then the power management software monitor 80 will force the system into the power conservation mode (DOZE or SLEEP) per the FIG. 6 operation regardless of the status of the hardware monitor 79. If the activity count is equal to or less than the active threshold, T_L, then the system will be programmed into the ON state.

[0102] The ON state can also be entered if the hardware monitor 79 detects a predetermined set of address ranges on bus 5. For example, the predetermined set of address ranges monitored for power management typically includes all of the I/O address range, that is, the addresses of the I/O controllers 6-0 through 6-n, and the video memory address range for the video memory locations 8 with the memory 11. Of course, other address ranges can be added to or used as the predetermined set for power management. The set of address ranges including the video memory and the I/O address range has been found to provide excellent information for controlling power management.

[0103] After entering the ON state, the power management unit will continue to be in the ON state until any idle function call detects the activity count has reached or gone beyond the idle threshold, T_H.

[0104] There are application programs such as Microsoft's Windows described in connection with FIG. 10 that do not use the DOS idle function calls and therefore the system would never go into the DOZE state through operation of the software monitor 80. Therefore, a watch dog timer is built into the power management software monitor to monitor the absence of idle function calls as indicated in connection with FIG. 7. If a time period greater than T_awas shown in the flow chart in FIG. 7 has been exceeded without any idle function call being made, then it is assumed that the application program bypasses DOS and goes directly to the hardware.

[0105] During the T_awtime period (see FIG. 7) the power management unit will be forced into the ON state until detection of activity for predetermined period of time, T_aw. This period, T_awis normally more than a minute in order not to affect the system performance. There is no power saving during the time out period, T_aw, even if the CPU is actually idling. After the T_awtime period, the hardware monitor 79 will take over completely.

[0106] In most cases, application programs go through DOS to perform I/O operations. The power management software monitor 80 keeps track of all the operating system function calls. If the accumulative count of all active and idle function calls is greater than the upper threshold, T_H, then the system is assumed to be inactive. The power management software monitor will program the power management unit to DOZE state only if the system is still in ON state. The computer 3 will enter DOZE state without waiting for the ON state timer to expire and therefore maximizes the power saving of the system. If computer 3 is already in DOZE or SLEEP, no action will be needed from the power management software monitor until the system becomes active again.

[0107] In the software monitor 80, inactivity is determined by detecting how many active or idle function calls an application makes within some time period. In the IBM PC DOS environment, the activity status is checked no less frequently than every 50 milliseconds. There are 256 IBM PC DOS function calls and each is labeled as idle or active with a corresponding positive or negative number. A positive number is assigned to an active function call and a negative number to an idle function call. The power management software module keeps a running total of the accumulated value of the function call numbers as the function calls are made. Whenever a function call is made, (either active or idle), the power management software module algebraically adds the number to the accumulated value and decides whether the system is active or not by comparing the magnitude of the accumulated value with a function call threshold. The function call threshold for determining activity is a variable depending on the computer system speed.

[0108] To prevent the system from oscillating between the active and idle state due to minor changes in system activity, hysterisis is provided by using active, T_L, and idle, T_H, function call thresholds. The accumulated total is clamped at T_Hafter it reaches the active thresholds T_Hor T_Las the case may be. The active and idle thresholds are typically unequal (128 and −256) so that the entry and exit from conservation (idle) mode is biased. For example, in order to have the system enter the idle mode quickly and thereby to reduce power consumption, the active threshold is set with a threshold number (128) greater than the idle threshold number (−256). Also, functions that require immediate attention are assigned numbers large relative to the active and idle thresholds so that a single occurrence of the function call (for example, I14=−400) will force the accumulated count over the active threshold (T_L=−256) and thus force the system to be in the active mode. The hysterisis effect can be bypassed by forcing the power management unit into active mode without changing the activity count. In this case, the next idle function call will bring the system back to idle mode.

[0109] If the software monitor 80 or the hardware monitor 79 indicates inactivity, the power management unit enters the conservation mode which has multiple states with different levels of power conservation.

[0110] The hardware monitor 79 works in conjunction with the software monitor 80 linked to the operating system during boot up time. The state control unit 23 is controlled by the timer unit 24 and power management software module 100. The power management software will override the hardware timer unit 24 whenever inactivity is detected in the operating system level. Since this can be done in a much finer resolution than the hardware monitor 79, the combined software and hardware monitor maximize power saving without any degradation in system performance.

[0111] Power Management Unit Detail—FIG. 3

[0112] Line List

[0113] In FIG. 3, the following lines and functions are defined for the connections output (O) from and input

[0114] Registers

[0115] In FIG. 3, the PMU 15 includes a number of registers accessed for read or write by CPU 4 over bus 5 via an index register addressing scheme. When not accessed by CPU 4, for example, after a power on detection by detector 71, the registers are all initialized to a default state. When accessed by CPU 4, an index value is first written to the index register 50 from bus 5 and the index value is decoded by decoder 70 to select one of the registers of PMU 15 for access to bus 5 to receive or send information from or to CPU 4. The index register 50, after an index write, is changed to point to another register to be accessed. When reset, the index register is not active to enable any PMU 15 register. This is a safety feature to help prevent applications executing on the CPU 4 from inadvertently accessing PMU 15 registers. All registers may be read and written over bus 5.

[0116] In register 51, only D0 and D1 are affected by a write. The CPU 4 can write the state code to this register to put the PMU in another state. Writing OFFh puts it in the OFF state. The NMI cause, state and wakeup codes are decoded as follows:

[0117] *RESET Bets STATE[0 . . . 1] and clears all other bits.

[0118] Supply Register

[0119] This register 52 is read only. D[0 . . . 2, 5] are driven directly by the input lines. Bit D3 is set when system activity is detected and is cleared when this register is read.

[0120] In register 53, the RING[0 . . . 2] bits are used to set the number of RI pulses required for turnon. The default value is 1 so that only one pulse is required for turnon. If set to 0, RI is disabled. State logic 23 has conventional logic for detecting and counting RI pulses from a modem, one of the I/O peripherals 7-0 to 7-n. D3 is only used for static CPU's. SLOW indicates reduced clock speed operation in On. The CCLK[0 . . . 1] bits select the clock divisor for CLKOUT in SLEEP and DOZE states, and in ON if SLOW is set, according to the table.

[0121] The activity monitor ACTIVITY output is the logical OR of all unmasked activity sources. This register 54 affects only the ACTIVITY output. Refresh DMA cycles (*DACK0 low), interrupts, or accesses to the PMU 15, never affect the activity monitor 16.

[0122] NMIMASK Register

[0123] This register 55 masks the various NMI sources. In the default state only the INMI input can generate NMI.

[0124] Referring to register 56, OSCDIV[0 . . . 3] plus one is the OSC frequency in MHz, except for OSCDIV[0 . . . 3]=13, the default, indicates 14.318 MHz. SLWREF is set when slow refresh DRAM is used. RASWIDTH[0 . . . 1] indicates the width of the *RAS pulse in units of OSC periods. The default value is 0 which disables refresh in SUSPEND state, and no RAS/CAS is generated. Values of 1 to 3 indicate 1 to 3 OSC periods.

[0125] PWR Registers

[0126] The bits D[0 . . . 7] in these registers 57 through 60 correspond directly with the power control outputs VP[0 . . . 7]. In a particular state, the corresponding PWR register outputs control the VP lines 23. The exception is VP0 and VP1 which are LCD and EL power, respectively. These outputs are AND'ed in AND gates 41 and 42 with the LCD and EL timer outputs prior to driving the lines 33. All bits are then exclusive NOR'ed in gates 35 with the POLARITY register 61, and the result drives the lines 33. The default values for these registers are as follows, where 1 indicates that the controlled device is on:

[0127] POLARITY Register

[0128] This register 61 controls the polarity of the VP outputs. If a logic low is required on a VP line to turn the external device on, the corresponding bit in the POLARITY register 61 must be low. If a high is required, set the bit high.

[0129] Timer Registers

[0130] The nonzero value loaded into one of the timer registers 63 through 68 is the actual timeout minus one. A zero disables the timeout. Therefore a 4 bit timer can be set for a timeout from 1 to 15 time units. Reading a timer register returns the value that was last written to it, not the actual time remaining. The default values are tabulated below:

[0131] OUTPUT Register

[0132] The OUTPUT register 62 is a read only register. For each VP[0 . . . 7] output that is on, the corresponding bit in the OUTPUT register will be set.

[0133] The control and logic functions for the activity monitor 16, the state logic 23, the NMI logic 21, and other components of FIG. 3 are conventional logic circuits for implementing the logic and control functions hereinafter described or alternatively are the software logic of TABLE 1.

[0134] ON State

[0135] Referring to FIG. 4, the ON state is entered from the SUSPEND or OFF state when the *RESET input is low, and also when one of EXT, RTC or RI goes high if ACPWR is true or LB is false. It is entered from DOZE or SLEEP when the activity monitor 16 detects activity with addresses in the predetermined address set. In the ON state encoded on lines 34, all power control outputs VP[0 . . . n] will be controlled by the PWRON register 57. Upon entering the ON state, the DOZE timeout timer 63 will be retriggered. The LCD and EL timeouts in timers 66 and 67 will be retriggered when entering the ON state from SUSPEND or OFF. The retrigger lines from STATE logic 23 to the timers are not shown in FIG. 3 for clarity.

[0136] In FIG. 3, the STATE logic 23 recieves the CPU data bus D(0 . . . 7) from bus 5 for receiving state commands issued by the software monitor 80 of TABLE 2. The STATE logic also receives the address detection line 76 from activity monitor 16 which enables the STATE logic 23 to receive the state commands from the software monitor when addressed over the bus 5.

[0137] If the SLOW bit in the control register 53 is false, the CLKOUT rate on line 28 will be full speed. If the SLOW bit is true, CLKOUT will be as specified by the CCLK[0,1] bits in register 53. This clock control allows the user to save power, for example, when running non-computationally intensive applications such as word processing.

[0138] DOZE State

[0139] The DOZE state is entered from the ON state when the activity monitor 16 has not detected activity and therefore has not provided the ACTIVITY signal within the time, T1, specified by the DOZE timer 63. In the DOZE state encoded on lines 34, the power control outputs VP[0 . . . 7] from unit 17 are controlled by the PWRDOZE register 58. If a non-static CPU 4 is used, the clock on line 28 will be slowed as specified by CCLK[0,1] in register 53.

[0140] If a static CPU 4 is used, CLKOUT on line 28 will stop in the low state immediately following a non-DMA memory read instruction, as indicated by *S1 going high while *AEN is low, so that no chip select will be low. If INTR goes high, CLKOUT will be enabled until after EOI is written to the interrupt controller with INTR false. If INMI goes high, CLKOUT will be enabled. If an internally generated NMI occurs, CLKOUT will be enabled until the NMIMASK register 55 is read. If any DRQ goes high, CLKOUT will be enabled until after the next memory read instruction with AEN and all DRQ inputs false. The enable request functions for INTR, INMI, internal NMI and DMA are separate and CLKOUT is enabled when any event requests it, so that an interrupt handler in CPU 4 will run to completion even if it is interrupted by a DMA request. These enable request functions are independent of the activity monitor and the ACTMASK register 54. Enabling CLKOUT does not cause the PMU 15 to leave DOZE, unless the activity monitor 16 is subsequently triggered. If this trigger occurs, the PMU 15 will enter the ON state and the enable request logic will be cleared.

[0141] SLEEP State

[0142] The SLEEP state is entered when the PMU 15 has been in the DOZE state for the time, T2, specified by the SLEEP timer 64 and no ACTIVITY signal has occurred. In the SLEEP state, the CLKOUT operation is the same as in DOZE. The power control outputs are controlled by the PWRSLEEP register 59.

[0143] Alternatively, the PMU can be programmed to generate NMI and remain in DOZE state instead of automatically entering SLEEP.

[0144] SUSPEND State

[0145] The SUSPEND state is entered when the PMU 15 has been in the SLEEP state for the time, T3, specified by the SUSPEND timer 65 or when a power check detects low battery signals, LB or LLB. The SUSPEND state is entered after these conditions only when the CPU 4 writes the code for SUSPEND to the STATUS register 40 and this operation requires software support because in SUSPEND the CPU operation is affected. In SUSPEND operation, CLKOUT is the same as CLKI. The power control outputs are controlled by the PWRSUSPEND register 60. In SUSPEND, the CPU 4 and the device (for example, a switch) which generates the system reset signal must be powered off. Only activity on the EXT, RI or RTC inputs can cause an exit from SUSPEND, and the new state after exit will be ON. When the reset circuit power is restored, it will reset the CPU 4, which will then execute a warm startup routine in a conventional manner. DRAM refresh may be enabled in SUSPEND. If DRAM refresh is not enabled, the PMU 15 does not need OSC from unit 43 in SUSPEND, and gates it off internally to minimize OSC power consumption. The OSC output will stay low. The bus interface is inhibited, and the data bus 5 is tristated.

[0146] OFF State

[0147] The OFF state is entered when the CPU 4 writes the code of OFF (OFFh) to the STATUS register 51. It is also entered 5 seconds after the EXT input goes high if the NMI is not serviced.

[0148] The OFF state is meaningful only when the PMU 15 is powered from a battery while the rest of the computer 3 is turned off. This type of power connection is necessary only if the PMU 15 must awaken the system from the OFF state by activating VP outputs on lines 33 in response to transitions on the EXT input. If this function is not required, then the PMU 15 may be powered off when the system is powered off, and the OFF state as described below is not required.

[0149] In the OFF state, all outputs from the PMU 15 are either low or tristated, and all devices other than PMU 15 in the computer 3 are powered off. Any inputs will have pulldowns so that floating inputs, if any, will not cause increased power dissipation. Only activity on the EXT, RI or RTC inputs can cause an exit from OFF, and the new state will be ON. The bus 5 interface is inhibited and data bus 5 is tristated.

[0150] Activity Monitor

[0151] The activity monitor 16 includes an address detector 73 which receives addresses from bus 5 representing the address activity of the CPU 4. The address detector 73 receives, for example, control lines and address lines SA(0 . . . 9) from bus 5 for sensing when those addresses are within the predetermined address set. The predetermined address set is defined, for example, by an address set specified by ACTMASK register 54. The detector 73 compares or masks the address set specified by register 74 with the addresses on bus 5 and provides an address detect signal on line 76 to the logic 77. The logic 77 receives the other inputs to the activity monitor 16 and combines them, using conventional logic circuitry, to provide three outputs.

[0152] The three outputs provided by activity monitor 16 are produced by conventional logic or by software as shown in TABLE 1. The EXTRIG output is a function of keyboard activity only and is used to retrigger the EL backlight timer 67. The LCDTRIG output is true for keyboard activity or video memory writes, and retriggers the LCD timer 66. The ACTIVITY output is an OR function of a programmable selection of different activities specified in the ACTMASK register 54. When active, this output returns the PMU 15 to the ON state and retriggers the DOZE timeout timer 63. The activity monitor 16 does not produce the ACTIVITY output in response to accesses to the registers of PMU 15.

[0153] OSC Programmability

[0154] The OSC frequency of refresh control unit 20 provides the timebase for the timers and the refresh for DRAM memory 11. The PMU 15 may be programmed to accept a range of OSC frequencies. The OSC frequency of oscillator 43 is fed to a counter 44 which divides it by a divisor which is programmed in the OSC register 56. The programmable counter output of divider 44 is divided to produce 256 Hz which is used by the refresh control logic 48. Further dividing in divider 46 produces 32 Hz for slow refresh to refresh control logic 48, and 8 Hz and 1/(7.5) Hz for use by the timers 63, 64, 65 and 68.

[0155] Timers

[0156] There are six timers in the PMU 15, namely, DOZE timer 63, SLEEP timer 64, LB (low battery) timer 68, SUSPEND timer 65, EL (backlight) timer 66, and LCD timer 67. Each of the six timers a 4-bit register loadable by CPU 4 over bus 5. Setting a timer register to 0 disables it; setting it to a nonzero value enables it. If enabled, certain timers are triggered by the transition to the ON state. Individual timers are also triggered by events specific to their functions. Some timers are retriggerable, timing out at a programmable time following the last trigger.

[0157] The DOZE timer 63 is programmable from 1 to 15 seconds with a resolution of 1 second, and the SUSPEND timer 65 is programmable from 5 to 75 minutes with a resolution of 5 minutes. All other timers are programmable from 1 to 15 minutes with a resolution of one minute. There is a quantization error associated with retriggering any timer. This error is a quantization error associated with retriggering any timer. This error will cause the actual timeout to be up to ⅛ of the resolution of the timer longer (but never shorter) than the programmed value. The error does not vary with the programmed value.

[0158] The LCD timer 66 and the EL timer 67 are retriggerable. The timer outputs are AND'ed in AND gates 41 and 42 with the power control bits selected by the power control multiplexer 76 according to the current PMU state to control the LCD (VP0) and EL (VP1) power control outputs to EXOR 35. This operation provides the flexibility to turn the EL and LCD outputs off when the associated timers 66 and 67 time out, or to control the outputs in any PMU power-management state under control of multiplexer 76.

[0159] The DOZE timer 63 is retriggerable and is triggered by the activity monitor ACTIVITY output in the ON state, and triggers the transition to DOZE state when it times out.

[0160] The SLEEP timer 64 is triggered when the DOZE state is entered and is cleared when the DOZE state is exited. Timer 64 either generates NMI or triggers the transition to SLEEP state when it times out.

[0161] The SUSPEND timer 65 is triggered when the SLEEP state is entered and is cleared when SLEEP is exited. If unmasked, an NMI will be generated when it times out.

[0162] The LB timer 68 is enabled when ACPWR is false (no AC power). Timer 68 is triggered when LB is first detected. If unmasked, NMI is generated by the LB timer 68 output once per minute when it times out, until a period of one minute elapses during which LB remains continuously false. The NMI cause will be identified as an LB or LLB interrupt. Software can maintain a counter and display a message once per X interrupts. It can also monitor LLB and shut the computer down after Y interrupts. It can also monitor LLB and shut the computer down after Y interrupts with LLB true.

[0163] NMI

[0164] The PMU unit 15 OR's together a number of internally generated NMI requests to produce the NMI output on line 27. These requests can be masked by bits in the NMIMASK register 55. The INMI input comes from conventional external NMI-generating logic such as a parity detector, and can be OR'ed with the internal NMI requests to generate NMI when unmasked by the OS2 bit in the NMIMASK register 55. The NMI output on line 27 generally goes to the CPU NMI input, except on OS2 systems where it must go to an IRQ. The NMI CAUSE code bits in the Status register 40 indicate the cause of the NMI on line 27. An internally generated NMI is cleared by reading the NMIMASK register 55.

[0165] NMI may be generated to indicate a low battery when ACPWR is false.

[0166] If the MSKSLEEP bit is cleared, the PMU 15 will generate NMI when the SLEEP timer 64 times out and remain in DOZE instead of entering SLEEP.

[0167] NMI is also generated when the SUSPEND timer 65 times out. Software can then save status and go to SUSPEND or OFF state.

[0168] A high on the EXT input while not in the OFF or SUSPEND state will generate NMI. Software can then save status and go to SUSPEND or OFF state. If the NMI is not serviced within 5 seconds, the PMU 15 assumes there is no software support for SUSPEND and will turn all power off and enter the OFF state.

[0169] Refresh in SUSPEND State

[0170] Refresh is enabled by setting the RASWIDTH[0 . . . 1] bits in the OSC register 56 to a nonzero value. This enables OSC to run in SUSPEND mode, and the RASWIDTH value also sets the width of the *RAS pulse in units of OSC clock periods. Slow refresh is enabled by setting SLWREF high. The PMU 15 generates *MRAS and *MCAS signals to mux 32 to refresh DRAM while the CPU is powered off or being reset. When the CPU is active, the *PCRAS, *PCCAS signals on lines 30 from the PC controller 13 are selected by multiplexer 30 to provide the *RAS, *CAS signals on lines 29. *REFRSEL on line 72 will go low to indicate that the PMU 15 is controlling refresh and high for PC controller 13 control.

[0171] If enabled, the DRAM refresh outputs are active in SUSPEND. When entering SUSPEND, the PMU 15 immediately generates a burst of 1024 CAS before RAS refresh cycles. A burst of 256 cycles is then repeated every 3.9 ms if SLOWREF is false or every 31.25 ms if SLOWREF is true. After entering the ON state from SUSPEND, the PMU 15 generates bursts of 1024 refresh cycles over 2.9 ms. This operation allows as much time as needed for CPU power stabilization, crystal oscillator startup and CPU reset. When the CPU is ready to take over control of the DRAM, it must poll the SUPPLY register 38 until the DRAMRDY bit goes high. The PMU 15 senses the polling operation as a request from the CPU for DRAM control, and at the end of the first refresh burst following a CPU I/O read of the SUPPLY register 38, the PMU 15 sets *REFRSEL high to return control of the DRAM to the CPU. The DRAMRDY bit is essentially the same signal as *REFRSEL.

[0172] The purpose of the bursts when entering and leaving SUSPEND is to eliminate violations of the refresh rate spec when switching between external refresh row address generation (DMA cycles during ON) and internal row address generation (CAS before RAS during SUSPEND).

[0173] Pseudostatic RAM refresh is also supported. When *REFRSEL goes low, *RAS can drive *RFSH low for auto refresh mode. The burst refresh will assure that switching between external and internal refresh will not violate the refresh rate spec. Self refresh can also be used by driving *RFSH low when *REFRSEL is low, but other logic will have to generate the refresh burst when entering and leaving SUSPEND, if required.

[0174] External Wakeup Inputs

[0175] RI is a rising edge sensitive input, to state logic 23 from a modem ring indicator RI output of a peripheral 7. The number of rising edges required for this input to be recognized is specified in bits D[4 . . . 6] of the Control register 53. The default is one transition. If these bits are zero, this input is disabled. If enabled, a rising transition on this input will force the PMU 15 to the ON state.

[0176] RTC is an edge sensitive wakeup-alarm input from a real time clock in CPU clock control 49 of FIG. 3. A rising or falling transition on this input will force the PMU 15 to the ON state.

[0177] EXT is a rising edge sensitive input, intended for use with an external pushbutton. A rising transition on this input while the PMU 15 is in OFF or SUSPEND will force the PMU 15 to the ON state. A transition in ON, DOZE or SLEEP will generate NMI.

[0178] EXT is debounced in ON, DOZE and SLEEP in a conventional debouncer circuit 36. A rising edge immediately generates NMI but only if EXT has been sampled low at least twice by a 32 Hz debounce clock from counter 46 prior to the rising edge. The debounce clock is derived from OSC 43 and therefore may be stopped in SUSPEND and OFF, so the PMU 15 will not enter these states until the debounce operation is completed. To prevent resuming due to contact bounce on the release of a pushbutton, the PMU 15 will defer execution of a change of state command from the CPU 4 until after the EXT input has been sampled low twice by the debounce circuit 36. This operation is typically transparent to software. For example, if the user presses the button in ON, the PMU 15 will generate NMI, and the CPU will write the command to enter SUSPEND and then execute a halt instruction. Nothing will happen until after the pushbutton is released, at which time the PMU 15 will enter SUSPEND.

[0179] Resume and Power on

[0180] The PMU 15 has its own private *RESET signal, typically from an external RC network detector 71 which detects VCC. This signal resets only the PMU 15 when power, VCC, is first applied to it. A separate reset signal must be generated by external hardware for the CPU when entering the ON state from SUSPEND or OFF state. At power on, the CPU 4 must read the RESUME bit in the Status register 51. RESUME will be cleared if the startup is a cold start from OFF and will be set to indicate a warm start (resume) from SUSPEND. If RESUME is cleared, the wakeup bits WU[0 . . . 1] in the Status register 51 will be zero, otherwise they will indicate which external input caused the resume. The RESUME bit will be cleared after the Status register is read.

[0181] Clock Switching

[0182] The clock switch control 69 is provided to switch input clocks CLK1IN and CLK2IN clocks to output clocks CLK1OUT AND CLK2OUT for peripherals. The CLK1 and CLK2 operations are the same. For example, the CLK1IN is passed to the CLK1OUT output by control 69 in ON and DOZE. When entering SLEEP mode, CLK1OUT will stop synchronously in the low state. CLK1OUT will start synchronously when returning to the ON state.

[0183] Low Battery Detection

[0184] The LB and LLB inputs indicate low battery and low low battery as generated by a conventional battery level detector in power supply 9 of FIG. 1. The polarity of these inputs is programmable by the LBPOL line which can be strapped low or high. If this line is high, LB and LLB are high true. If low, these inputs are low true. The status of the LB and LLB lines after polarity correction can be read in the SUPPLY register 38. A low battery indication can generate NMI.

[0185] Power Sequencing

[0186] To minimize turnon transients, the turnon of VP1 (EL power) is delayed by 4 to 8 ms after OSC begins clocking, when entering the ON state.

[0187] Program Listing

[0188] A computer program embodiment of the hardware monitor for the power management unit appears in the following TABLE 1.

[0189] While the invention has been particularly shown and described with reference to preferred embodiments thereof, it will be understood by those skilled in the art that the foregoing and other changes in form and details may be made therein without departing from the spirit and scope of the invention.